Frequently Asked Questions

A desktop coding app with AI built in. You open a project, describe what you want, and a team of AI agents plans the change, writes it, reviews it, checks it for security problems and tests it. You see the result as a diff, and nothing is written to your files until you click Apply. Codira also has a normal editor, a terminal, git tools and a live preview of your app, so you can do the whole job in one window.

In short: "Expert engineers + agentic AI collaborate to build better software." AI writes code quickly but makes mistakes. Codira's answer is to put every change through several separate checks before you see it, and to keep a person in charge of what gets applied. In practice that means:

• Specialist agents. Planner, Implementer, Reviewer, Security, QA, Auditor, Verifier, UX Reviewer, Explainer and Debugger.
• Six patch guards. Rule-based checks that use no AI. They catch unexpected deletions, references to files or symbols that don't exist, needless rewrites, placeholder code, dropped behaviour and design-system violations.
• Verification after you apply. Codira builds the project, runs your tests, smoke-checks the running app, and the Verifier reports what actually passed.
• An undo for every AI change. Time Machine saves a checkpoint before each apply.

No. Codira has two modes for plain chat requests.

• Freestyle is the default. One fast agent writes the change, and after you apply it Codira builds, fixes any build errors and runs your tests. It skips the Security and QA agents and the independent Verifier to stay quick, then offers Harden with Orchestrate to run those checks afterwards.
• Orchestrate runs the full pipeline: Planner, Implementer, Reviewer, Security and QA, then the build check, tests, a smoke check of the running app and the Verifier.

You choose the mode in Settings → AI Models. ⌘K Composer and /plan have their own fast and full-review options; see How Codira's AI works and Chat, Composer, completions and commands.

Single changes wait for you to click Apply. Two exceptions are worth knowing:

• When you approve a multi-step /plan in chat, each step is applied automatically once the Reviewer approves it, so the next step can build on it. Each step still gets its own Time Machine checkpoint.
• Codira's own build, test and install steps run after you apply. Anything that installs, deletes or pushes asks you first.

Codira, Inc. The website is codira.com, and apps you host with Codira live at codira.space.

• Bryan Ferre, Founder and CEO
• James K. Farrelly, President and COO
• Christian Brotherson, Chief Product Officer
• Robert Clawson, Business Development and Investor Relations

Mainly working developers who ship real code, solo founders, and teams that want AI changes to meet a consistent standard. You get the most from it if you can read a diff and decide whether a change is right. Codira also has pages for:

• Vibe coders and non-engineers. Sidekick, a plain-English helper, turns what you want into a precise request for the agents, and Genesis and the Setup Wizard can start a project from a one-sentence idea.
• SaaS companies and software agencies. Fast onboarding to unfamiliar codebases with /explain, team standards, and checks that catch problems before a demo.
• Small businesses. Building your own internal tools. These are things you build with Codira, not ready-made features.

• The desktop app for Mac and Windows (see Download, install and updates). This is where you edit, run the agents, preview your app and use git.
• Your codira.com dashboard. Your plan, credits, invoices, settings, team workspace, projects board, hosting and bug reports.
• Hosting. Press Launch in the app and your project goes live at a yourname.codira.space address. Paid plans include one Small app. See Codira Hosting.

Neither. It's built with Tauri (a Rust desktop shell) and the system web view. The code editor inside it is Monaco, the same editor engine VS Code uses.

No. There's no extension marketplace or plugin system.

Yes. They can open the same project folder at the same time.

Codira can send each change through several specialist agents and six rule-based guards before you see it, gives you a single "Ready to apply / Apply with care / Don't apply yet" recommendation, and then builds, tests and checks the running app after you apply. Every AI change is checkpointed, so you can undo it in one click. Codira also has autocomplete and a fast single-agent mode (Freestyle) when you want speed.

Codira works on a real project folder on your computer, in any framework. With the visual editor you click an element in the live preview and change it directly, and the edit goes into your actual source file instead of a re-generated page.

No. It uses models from Anthropic (Claude), OpenAI, Google (Gemini), xAI (Grok) and DeepSeek. You can pay for them with Codira credits (see Credits and usage) or use your own API keys. See AI models, providers and your own keys.

No. The core is closed source.

Partly. The editor, terminal, file browser, git, Time Machine and previews of a local dev server all work offline. AI features need an internet connection. See Accounts and signing in for how long the app trusts your plan while offline.

• codira.com/tour is a slide-deck tour. You can link to a specific slide with ?slide= and a number.
• codira.com/demo is a form to request a walkthrough. Someone replies within one business day.
• Inside the app, a short feature tour runs the first time you open a project. See Getting started.

• Mac: macOS 11 Big Sur or later, on Apple Silicon or Intel.
• Windows: Windows 10 or 11, 64-bit.
• Linux: not available. It's on the roadmap with no date.
• Chromebook: not available.

Yes. Since version 0.12.16 the Mac app is a universal binary, with Intel and Apple Silicon code in one app. Your Mac runs the right one automatically, so Apple Silicon doesn't need Rosetta.

No. You need the desktop app.

• git is required. Codira checks for it at launch.
• Node.js and npm for previews and JavaScript builds.
• The GitHub CLI (`gh`) if you want to create GitHub repos or open pull requests from Codira.

Codira helps you install these when a feature first needs them. See Getting started.

Click Download on codira.com, which goes to codira.com/get. Pick macOS or Windows, then enter your name, email and a password (at least 8 characters) to create your free account. The installer downloads straight away, and we email you a 6-digit code to confirm your email.

From Codira's public release page on GitHub (Codira-Software/releases). codira.com/download sends you to the newest file there. A new release appears on the site within about 5 minutes of publishing.

The Mac download is 1.6.9. The newest Windows installer is 1.6.7, because the two platforms don't always ship together. Settings → About Codira shows the version you're running.

1. Open the downloaded DMG.
2. Drag Codira into Applications.
3. Open Codira from Applications or Spotlight.

Codira notices and offers to copy itself to Applications and relaunch. Running from the DMG stops updates working and stops you ejecting the disk image, so say yes.

Yes. Every release since 1.0.0 (7 June 2026) is signed with an Apple Developer ID and notarised. The notarisation ticket is attached to the app, so the first launch works even without an internet connection.

Yes. macOS asks this the first time you open any app downloaded from the internet. Click Open.

A current download shouldn't do this. Delete the copy you have and download a fresh one from codira.com.

No. It runs with Apple's Hardened Runtime instead, and limits itself: it only reads and writes files inside the project folder you open, and it checks every command it runs for you (see Safety, review and undo).

Run the …_x64-setup.exe installer. It installs for your user only, so you don't need admin rights. It needs Microsoft's WebView2 runtime, which Windows 11 and current Windows 10 already have; if it's missing, the installer downloads it.

• The built-in terminal is PowerShell.
• Keyboard shortcuts use Ctrl where the Mac uses ⌘.
• Automatic updates aren't set up on Windows yet. Download new versions by hand.

Yes, on Mac. It checks when it starts and every 6 hours while it's running. When there's an update, a banner appears in the bottom right with Install and Later. Install downloads the update, checks its signature, installs it and restarts Codira. Later hides the banner until the next launch. Nothing interrupts your work.

Yes: Settings → About Codira → Check for updates. If you're up to date it says "You're on the latest version".

Two separate signatures protect it. The app itself is signed and notarised by Apple. Each update package is also signed with Codira's own update key, and the app refuses any update whose signature doesn't match the key built into it.

Versions 1.6.3 to 1.6.6 weren't offered updates because of a problem in our release process, now fixed. If you're on one of those, download the latest version from codira.com once, and updates will work from then on.

Often, usually several times a month. Help → What's New in the app opens the changelog.

• 0.12.16: Intel Macs supported (universal app).
• 1.0.0 (7 June 2026): first notarised public release.
• 1.3.x: build and test checks for more languages, and the Verifier.
• 1.5.4 (22 July 2026): a security rebuild of 1.5.3, made from verified-clean source with new signing credentials after a problem in the build configuration was fixed. Anyone on 1.5.3 or earlier was asked to download it.
• 1.5.5 (23 July 2026): first Windows installer.
• 1.6.x: Codira hosting, git sync improvements, staying signed in between launches, the prerequisites helper, the iCloud/Dropbox warning and the reworked Setup Wizard.

~/Library/Application Support/dev.codira.ide/settings.json

Preferences, window position, open tabs, last project, cached plan. Only your user account can read it.

~/Library/Application Support/dev.codira.ide/conversations/

Chat history, one file per project

macOS Keychain, items named dev.codira.ide:…

Your API keys, sign-in token, Jira token and per-project environment variables

.codira/ inside each project

Project memory, standards, Architect data, search index and other project files (see Getting started)

Hidden git refs inside each project

Time Machine checkpoints

Codira doesn't write a log file.

1. Quit Codira (⌘Q).
2. Drag Codira from Applications to the Trash.
3. To remove settings and chat history, delete ~/Library/Application Support/dev.codira.ide.
4. To remove saved keys and your sign-in, open Keychain Access, search for dev.codira.ide and delete the items you find (:ai, :auth, :jira and any :env: entries).
5. Optional: delete the .codira folder in each project. To remove Time Machine checkpoints from a project, run this in its folder: git for-each-ref --format='%(refname)' refs/codira/ | xargs -n1 git update-ref -d

You may also find small cache folders named dev.codira.ide under ~/Library/Caches and ~/Library/WebKit, which you can delete.

Codira installs per user with a standard installer, so remove it from Settings → Apps.

At codira.com/signup (or codira.com/get when you download). Enter your name, email and a password of at least 8 characters. You can also add a phone number; if you do, a separate box lets you agree to SMS messages. You're signed in straight away.

No. Accounts use an email address and password.

After you sign up we email you a 6-digit code. You can use the website before confirming, but two things wait until you do: signing in to the desktop app, and starting your free trial. The codira.com dashboard also asks you to confirm first.

Click resend to get a new code. You can ask for one every 60 seconds, and each new code replaces the last.

Yes. Confirming your email starts a 7-day trial of Individual Core with a one-time 1,000 credits. No card is needed and you're never charged automatically. See Plans and pricing.

One trial per account and one per real email inbox. For Gmail addresses, dots and "+tags" are ignored, so j.smith+test@gmail.com counts as jsmith@gmail.com. You can't get the trial if your account has ever had a subscription, including a cancelled one.

You move to the free Hobby plan automatically. Nothing is deleted: your projects, chat history and checkpoints stay. Pick a plan any time from your dashboard or Settings → Account & Subscription in the app (plans and prices are in Plans and pricing).

No. Hosting needs a paid plan. See Codira Hosting.

At least 8 characters.

Click "Forgot password?" on the sign-in page and enter your email. If there's an account for it, we email a reset link. The link works once and lasts 1 hour, and asking for a new one cancels older links. You can request up to 5 links in 15 minutes. For privacy, the page shows the same message whether or not the email has an account.

Dashboard → Settings → change password. You'll need your current password.

Not yourself. Email support@codira.com. You can change your name, handle, bio and profile picture yourself.

No, not for customer accounts.

30 days, unless you sign out.

Through your browser.

1. Click Sign in in the app. Codira opens codira.com in your default browser.
2. Sign in there if you aren't already, and confirm your email if you haven't.
3. You see "Codira wants access to your Codira account" with the permissions listed. Click Authorize.
4. The browser hands you back to the app and shows a page saying you can close the tab.

The app waits up to 5 minutes for you to finish. It never sees your password.

Your profile (name and picture), your email address, your subscription (plan, status and features), your credit wallet (check your balance, set credits aside during a run, and charge for it when it finishes) and staying signed in.

Yes. Sign out, then sign in again. The browser asks which account to use instead of reusing the last one.

No. Since September 2026 the app keeps your sign-in in the macOS Keychain and renews it quietly each time it starts. Your sign-in stays valid for 30 days after the app last renewed it.

Most often your sign-in expired after a long break. It also happens as a safety measure: if the same saved sign-in is ever used twice (for example, a copied settings folder on a second Mac), Codira treats it as possibly stolen and signs that app out everywhere for your account. Just sign in again.

Yes, except for AI. The app remembers your last confirmed plan and trusts it for up to 7 days without a connection. After that, AI features lock with "Offline too long — connect to verify your subscription" until you reconnect. Everything else keeps working.

Yes. Sign in on each one separately. Settings stay on each machine and don't sync between them.

Each Codira account keeps its own agent settings, model choices, agent nicknames and Jira connection. Window layout, theme and tour progress are shared by everyone on that Mac user account.

Settings → Account & Subscription → Sign out. Codira asks whether to save any unsaved files, closes the project and forgets your plan details. Next time it opens to a clean start screen.

Use the sign-out button in the dashboard. It ends that web session.

It means the app can't run agents right now, usually because you're signed out. Click it to go to Settings → Account & Subscription.

Email support@codira.com or privacy@codira.com. There's no self-serve delete button. Cancel any paid subscription first.

There's no export button. Your code, chat history and project files are already on your own computer. Invoices are in Dashboard → Subscription, and your credit history is in Dashboard → Credits (see The codira.com dashboard). For anything else, email privacy@codira.com.

Codira can suspend an account, for example for a breach of the Terms or the Acceptable Use Policy. A suspended account can't use the desktop app's AI features. Email support@codira.com to find out why and what to do next.

A three-step welcome window:

1. Welcome: what Codira is.
2. Connect: sign in with your Codira account.
3. Get going: open a recent project, start a new project, or take the tour.

Codira also checks that git is installed. Then a short feature tour runs the first time you open a project.

Yes. Settings → About Codira → Show welcome tour. Each new project you open also gets a small "First time in this project? Take the tour" prompt.

• git: checked every time Codira starts, because a lot breaks quietly without it.
• Node.js and npm: checked when you first preview or build a JavaScript project.
• GitHub CLI (`gh`): checked when you first create a GitHub repo or pull request.
• Homebrew: not required. Codira only checks for it so its advice fits your Mac.

It helps, but it never installs anything silently.

• Apple's developer tools (which include git): a Run this for me button starts Apple's own installer.
• git or `gh` with Homebrew: a button types brew install … into the terminal so you can watch it.
• Node.js, or no Homebrew: Codira opens the official download page.

After installing, quit and reopen Codira, because a running app can't see newly installed tools. Codira checks for Apple's developer tools without triggering macOS's "Install Command Line Tools?" pop-up.

Codira reads your PATH the way a login shell does, which loads ~/.zprofile but not ~/.zshrc. If your Homebrew or nvm setup is only in .zshrc, move the brew shellenv (or nvm) line into .zprofile, then restart Codira.

Explorer → Open Folder, then choose the folder. Codira works on one folder at a time (your "workspace"). Next time you start Codira, it reopens your last project and tabs, as long as you're still signed in and the folder still exists.

No. Codira has one window with one project. To switch, close the folder (⌘⇧W) and open another.

Codira stops its dev server, closes the editor tabs, clears the chat and canvas, and saves that project's chat history so it's there next time.

Yes. Explorer → Clone Repository. Paste an HTTPS or SSH address and choose where to put it (by default a Codira folder in your home folder). Codira uses your own git, so private repos work with whatever you already use: the macOS Keychain, an SSH key or gh auth login. It never asks for a password; if git has no credentials, the clone fails straight away and suggests SSH or gh auth login. A failed clone is cleaned up. There's no progress bar, so a large repo can take a while. The cloned folder opens as your project.

Click Create New Project in the empty Explorer. Pick a framework and a name, then click Create. Codira runs the framework's official setup command in the terminal so you can watch, waits for dependencies to install, and opens the new folder. It usually takes 60 to 120 seconds. New projects go in a Projects folder in your home folder by default.

• Next.js (the default and recommended): TypeScript, Tailwind and the App Router
• Vite + React (TypeScript)
• Expo (dependencies aren't installed automatically)
• SvelteKit
• Astro
• Remix

All are JavaScript or TypeScript. For Python, Go, Rust or anything else, create the project your usual way and open the folder; Codira works with any text-based codebase.

Yes. Codira sets the project up in a temporary subfolder, then moves the files up into your folder. It never overwrites a file that already exists; any clashes are left in the subfolder and listed for you.

A guided path for a brand-new app, in seven steps:

1. What are we building?
2. Set up the project
3. Design system
4. Your brief
5. Build it
6. Project plan
7. Get set up to build

You can leave and come back; it works out which steps are done from what's in the folder. The "Build it" step creates a first version of the app and only works with Next.js.

For Node and Next.js projects that use Postgres, the "Get set up to build" step can start a local Postgres database in Docker (OrbStack is recommended on Mac), add its settings to .env.local without overwriting what's there, install a test runner and apply your schema. Each step that changes your system asks first. Nothing is created in the cloud and nothing is billed. Only Postgres is supported.

Sync services fight with build tools and create duplicate "file 2.ts" copies, even inside .git. Codira spots iCloud Desktop & Documents and any service under ~/Library/CloudStorage (Dropbox, Google Drive, OneDrive and others) and suggests moving the project to ~/Projects. It never moves or deletes anything itself, and you can dismiss the warning for each project.

Codira's notes and settings for that project. It's created the first time Codira needs to save something. You may see:

MEMORY.md

Facts every agent reads (add with /remember)

standards.md

Your team's house rules (/standards)

understanding.md

Codira's summary of your repo (/understand)

blueprint.md, genesis.md, backlog.json

Briefs and starter stories from Brainstorm and Genesis

architect.json

Your Architect stack and schema

settings.json

Shared model settings for this project

assets.json

The Asset Hub list of images and logos

embeddings.json

The search index

baseline.json

The style Codira learned from your code

plan-state.json, prerequisites.json, transcripts/

Working state

Commit the files your team should share, such as MEMORY.md, standards.md, understanding.md, blueprint.md, architect.json and settings.json. Codira adds the machine-specific ones (such as baseline.json, embeddings.json and transcripts/) to your .gitignore for you. You can delete the folder at any time; Codira rebuilds what it needs, though you'd lose your memory and standards.

Open any file, press ⌘K, type "add a 2-line comment at the top explaining what this file does", and run it. The agents plan and write the change and show you a diff. Click Apply to keep it or Reject to throw it away. ⌘K needs you signed in and a file open.

⌘↩ runs the full review (plan, write, review, security and QA). ⇧↩ runs fast mode (plan and write only).

• ⌘K Composer: edits to the open file or a selection.
• Chat: describe what you want. In Freestyle mode (the default) one agent does it.
• `/plan`: bigger features across many files, with the full agent pipeline.
• `/fix`: a quick, small change.
• `/explain`: questions about your code. On its own, /explain gives a five-part tour of your codebase: stack, architecture, entry points, conventions and suggested questions.

Chat can answer questions without a folder open, but it needs a folder to build anything.

Run /understand once. Codira reads the whole repo and writes .codira/understanding.md, which every agent then reads. /analyze re-runs a quick first-look scan, which is handy after pulling a lot of new code. Both use credits.

Monaco, the editor engine behind VS Code, built into the app. It uses JetBrains Mono at 13 pt, word wrap on, 2-space tabs, coloured bracket pairs and guides, and no minimap. Monaco's own features work as usual, including multiple cursors and find and replace within a file.

TypeScript and TSX, JavaScript and JSX, JSON, Markdown, CSS, SCSS, HTML, Rust, Python, Go, Java, C, C++, YAML, TOML, shell scripts and SQL. Other files open as plain text, including Swift, Kotlin, Ruby, PHP, C#, Vue and Svelte.

On purpose. The editor doesn't load your project's full setup (tsconfig, node_modules), so its TypeScript checking would mark correct code as wrong. It only shows syntax errors. For full type checking, run your project's type check in the terminal, for example npm run typecheck. After AI changes, Codira's build check does this for you.

No. Codira doesn't run a language server.

No. Press ⌘S to save. Unsaved tabs show a dot. Saves are written safely (to a temporary file, then swapped in), so a crash can't leave a half-written file.

No. The editor opens text files only. To add images to your project, use Import image in the Explorer.

Yes, as grey "ghost text" you accept with Tab. It's off by default. Turn it on with the AI Autocomplete pill in the bottom panel. See Chat, Composer, completions and commands for cost and limits.

Yes. Click Blame at the bottom right of the editor to show who last changed each line and how long ago. Hover a line for the full commit: ID, author email, date and summary. Lines you've changed but not committed show as "Uncommitted". Blame is switched on per tab. It doesn't follow renamed files or jump to the commit, and it can take a second or more on files with over 1,000 commits.

No. The font size is fixed at 13 and the shortcuts can't be changed.

Each file opens in a tab. ⌘W closes the current tab. When you start Codira, it reopens the tabs from your last session.

• Toolbar: new file, new folder, import image, refresh, close folder, open folder, collapse the panel.
• Right-click a file or folder: New File, New Folder, Rename, Delete.

Hidden files such as .env are shown; .git and .DS_Store are not. The tree updates when files change on disk.

No. Delete is permanent. Files are removed straight away, and folders with everything in them. Codira won't delete the project folder itself. Time Machine can only bring a file back if it was saved in an AI checkpoint, so commit your work often.

Click Import image in the Explorer and pick PNG, JPG, GIF, SVG, WebP, AVIF or ICO files. Codira copies them into public/ (or src/assets/) with web-safe names. Run /assets to add them to the Asset Hub, so the agents know what's there and can refer to "the logo" by name.

Press ⌘P and start typing part of the name. Use ↑ and ↓ to move, ↩ to open and Esc to close. It shows the top 30 matches, searches up to 10,000 files and skips folders like node_modules, .git and dist. It works once a folder is open.

Not yet. ⌘P opens files only. There's no command search, @symbol search or :line jump.

Yes, in the Search panel (the magnifier icon). Toggle case-sensitive (Aa), whole word (ab) and regular expressions (.*). Results are grouped by file.

Yes. Type a replacement in the Search panel. Regular expressions can use $1-style back-references. Codira shows every change as old → new, then asks you to confirm Replace All. Each file is re-read just before it's changed and saved safely. You can't skip individual matches, and there are no include or exclude filters.

Yes. Project search looks at up to 5,000 files, skips node_modules, .git, dist, .next and binary files, and shows up to 200 results. On very large repos some matches won't be counted.

Press ⌘⇧R. It fills in the selected text or the word at your cursor. Type the new name and you'll see every whole-word, case-sensitive match, grouped by file. Untick any files you want to leave alone, then press ⌘↩. Open tabs reload afterwards.

It's a text rename, not a code-aware one: it will also rename matching words inside strings, JSON and .env keys, and it doesn't understand scope. There's no undo button, so commit first (Time Machine doesn't cover it).

Yes, in the Terminal tab of the bottom panel. It's a real shell that starts in your project folder.

• Mac: your usual login shell (zsh by default), so your Homebrew and nvm setup loads.
• Windows: PowerShell.

It restarts when you switch projects.

No. There's one terminal per window. Switching to the Canvas tab and back keeps its history.

No. The terminal is yours, like the Terminal app. Only the commands Codira runs for you (builds, tests, installs and anything the AI suggests) go through its safety checks.

Some buttons run their command in the visible terminal so you can watch it and press Ctrl+C if you need to, for example "Initialize repository" (git init), brew install gh or installing the Vercel CLI. Codira clears anything you'd half-typed first.

Open the Canvas tab in the bottom panel. Codira works out your framework and starts your dev server, or connects to one you've already started in the terminal. The preview opens in its own window, which you can resize. Click ↗ Browser to open it in your normal browser with full developer tools.

Next.js (port 3000), Vite (5173), SvelteKit (5173), Astro (4321), Vue CLI (8080), Create React App (3000) and Expo (8081, running the web version). Codira looks in the project folder first, then in subfolders such as frontend/, web/, app/ and client/, and one level into apps/* and packages/*.

Desktop (1280 × 800), iPhone (393 × 852), Android (412 × 915) and iPad (820 × 1180). Expo projects open as iPhone; everything else opens as Desktop.

Preview Doctor checks before starting and explains failures with a one-click fix:

• Missing dependencies: installs them with npm, yarn or pnpm.
• Port already in use: Free port, which names the program using it and stops it.
• OpenSSL "digital envelope" error: retries with Node's legacy setting.
• A dev server already running in the terminal: the preview attaches to it instead of starting a second one.

Ordinary previews don't. Two optional features do:

• Turning on the canvas inspector for a Vite project installs Codira's small dev plugin and updates vite.config.
• Precise click-to-edit on Next.js adds a development-only helper to next.config.

No. The preview only shows your app running on your own computer (localhost). To visit other sites, use ↗ Browser, which opens your normal browser.

⌘S

Save

⌘W

Close the current tab

⌘⇧W

Close the project folder

⌘P

Open a file by name

⌘⇧R

Rename a word across the project

⌘K

Open Composer on the current file or selection

⌘↩ (in Composer)

Run with full review

⇧↩ (in Composer)

Run in fast mode

⌘↩ (in Source Control)

Commit

⌘↩ (in Rename and some dialogs)

Apply or submit

⌘B

Show or hide the side panel

⌘,

Open Settings

Esc

Close a dialog

⌘Q

Quit

Monaco's standard editing shortcuts also work in the editor. On Windows, use Ctrl in place of ⌘.

Explorer (folder), Search (magnifier), Source Control (branch), Agents (sparkles), Time Machine (clock), Ship (rocket), Architect (network) and Settings (gear). There are also icons for Sidekick (wand) and Design (pen nib).

Open it with ⌘, or the gear icon.

Account & Subscription

Your account and plan, credit balance, buying credits, auto top-up (paid plans only), sign out

API Keys

Your own provider keys, and how each provider is routed

Env Vars (per workspace)

Environment variables for this project, kept in the Keychain and written to .env.local

Jira

Your Jira Cloud site, email and API token

AI Models

Default model per provider, model per agent, agent nicknames, Execution mode (Freestyle or Orchestrate), chat formatting (formatted or raw), Verify build after apply, automatic UAT check after apply, and Use current framework versions

This Project

Shared model settings saved to .codira/settings.json

Team Standards

Your project's standards.md

Personal Style

Five personal preferences for how the agents write code

Appearance

Dark or Light theme

About Codira

Version, Check for updates, Show welcome tour, Send anonymous crash reports, and links to the docs, changelog and privacy policy

Dark (the default) and Light, under Settings → Appearance. The editor and terminal colours change with it.

Yes. Settings → This Project saves model choices to .codira/settings.json, which you can commit. Your licence, keys and theme are never included. Choosing a different model per agent for the whole project is done by editing that file by hand.

No. They stay on each machine.

The window opens at 1440 × 900 and can't be made smaller than 960 × 600. Codira remembers its size and position. You can collapse the side panel (⌘B) and the chat panel.

Yes: Codira (About, Hide, Quit), Edit (Undo, Redo, Cut, Copy, Paste, Select All), View (Full Screen), Window (Minimise, Close) and Help. Help has Codira Documentation, What's New, Report an Issue and Visit codira.com, which open the website in your browser.

Yes, in two places.

• Source Control panel (branch icon): see changed files, stage and unstage them (one at a time or all), write a message and commit with ⌘↩, and click a file to see what changed.
• Ship panel (rocket icon): commit and push, sync with the remote, branches, pull requests and creating a GitHub repo. It also handles deploying (see Shipping and deploying and Codira Hosting).

Yes. Codira checks for it at launch and helps you install it if it's missing (see Getting started).

Yes. Click a changed file in Source Control to see your version against the last commit, side by side or inline. New files that git isn't tracking yet can't be opened this way, and binary files show an error.

No. There's no discard button, no staging of individual changes inside a file and no compare against an older commit. Use the terminal for those.

It stages the files you picked (or everything), commits with your message and pushes. A new branch's first push also sets it to track the remote branch. If something fails, Codira tells you which step failed (stage, commit or push) and how to fix it.

Yes. Choose Amend in the Ship panel. It replaces your last commit and force-pushes safely, which means the push is refused if someone else has pushed to that branch since you last fetched.

Whatever your computer already uses. Push and pull use your own git with your existing credentials (Keychain, SSH key or gh). Creating repos and pull requests uses the GitHub CLI. The Ship panel shows "Connected as" with your GitHub name and warns you if more than one GitHub account is signed in to gh.

Run gh auth login (or gh auth switch) in the terminal. Codira deliberately has no switch button, because switching changes the GitHub account for every app on your Mac.

It fetches in the background when you open the Ship panel, when the Codira window comes to the front, and every 60 seconds, but never more than once every 30 seconds. Each fetch gives up after 20 seconds and never asks for a password. When your branch is behind or ahead, you see ↓ and ↑ counts and a Sync or Sync & Push button. If you're offline, the counts simply stop updating; there's no offline warning.

It pulls the remote's new commits and replays your local commits on top of them (git pull --rebase --autostash). Any uncommitted changes are set aside first and put back afterwards.

The Ship panel switches to a "stopped sync" view, and conflicted files show a red !.

1. Open each conflicted file and fix the sections between the <<<<<<< and >>>>>>> markers in the editor.
2. Click Finish sync to continue. It stages only the conflicted files you fixed.

Or click Undo sync to put everything back exactly as it was before you synced. Codira refuses to commit any file that still has conflict markers in it, and it also catches the case where your set-aside changes conflict even though git said the sync worked.

No. You fix conflicts in the normal editor. There's also no "skip this commit" option, on purpose.

Yes, in the Ship panel: see your local branches, create a new branch and switch to it, or switch to an existing one. Branch names are checked before use. You can't delete, merge or rename branches from Codira; use the terminal.

Yes. In the Ship panel, choose the base branch, enter a title and description, and optionally mark it as a draft. Codira pushes your branch and opens the pull request on GitHub. Once it exists, the panel shows its number, link and status for your current branch. You need gh installed and signed in.

Yes. Choose a name, public or private, and your personal account or any organisation you belong to. Codira creates the repo, connects your project to it and pushes.

Source Control and Time Machine show an Initialize repository button, which runs git init in the terminal. Codira also sets up git by itself in two cases: the first time you apply an AI change (Time Machine needs git for its checkpoints) and when you create a GitHub repo. In those cases it makes a first commit of everything in the folder, called "Initial commit (Codira baseline)". Files listed in .gitignore are left out, so check your .gitignore before your first AI apply.

Not by itself. AI changes are not committed: Time Machine saves its checkpoints as hidden git references that don't move your branch, don't appear in git log and aren't pushed. The only commits on your branches are ones you make, plus the one-off "Initial commit (Codira baseline)" in a folder that had no git. Amend only rewrites a commit when you choose it.

Yes, two things:

• The Ship panel adds a marked block for Codira's machine-specific files, such as .codira/baseline.json, .codira/embeddings.json and .codira/transcripts/.
• The Env Vars settings tab adds .env.local.

Files your team should share, such as standards.md and understanding.md, stay tracked.

Partly. Commit, push, fetch and sync work with any git remote. Creating repos, pull requests and the "Connected as" check work with GitHub only. There's no GitLab or Bitbucket integration.

The first git check loads your shell setup to find git and gh, which can take 0.5 to 1.5 seconds. After that it's cached and checks take about 50 ms. Codira starts this in the background when a project opens.

Five: Anthropic (Claude), OpenAI (GPT and the o-series), Google (Gemini), xAI (Grok) and DeepSeek. Codira doesn't make its own models. It sends each request to one of these providers.

Anthropic

Claude Opus 4.8, Claude Sonnet 4.6, Claude Haiku 4.5

OpenAI

GPT-4o, GPT-4.1, GPT-4o mini, o3 (reasoning), o4-mini (reasoning)

Google

Gemini 2.5 Pro, Gemini 2.5 Flash

xAI

Grok 4.20 (reasoning), Grok 4.3

DeepSeek

DeepSeek V4 Pro, DeepSeek V4 Flash

Claude Opus 4.8, GPT-4o, Gemini 2.5 Flash, Grok 4.3 and DeepSeek V4 Flash. You change these under Settings → AI Models. Changes apply from your next message, with no restart.

Every agent runs on Claude by default:

• Planner, Implementer and Reviewer: Claude Opus 4.8.
• Security, QA and UX Reviewer: Claude Sonnet 4.6.

These six are the agents you can route to a model of your choice. The other agents (Auditor, Verifier, Explainer and Debugger) can't be set separately.

Yes. In Settings → AI Models you can set any provider and model for each of the six agent slots. When more than one setting applies, Codira uses the first one it finds in this order:

1. A one-off choice in the ⌘K Composer for that run.
2. The project's shared settings in .codira/settings.json.
3. Your own per-agent choice.
4. Codira's default for that agent.
5. Your default model for that provider.

Yes. The Composer has a "Models" picker, collapsed by default. Expand it to change the provider and model for any slot for this run only. The chip shows how many overrides you've set. The choices lock once the run starts and are forgotten when you close the Composer.

Yes. Settings → This Project writes .codira/settings.json, which you commit to the repo. It can set a model per provider and a model per agent. The per-agent part has no visual editor yet, so you edit that file by hand. Project settings override each person's own choices. If a teammate's plan doesn't include a pinned model, Codira uses the closest model their plan does include. Your plan, keys and theme are never shared this way.

Yes. You can give any agent a nickname, for example "Sage" for the Planner. Nicknames only change what you see in the app. Your usage history still shows the real role names.

Codira groups models by cost.

• Standard models work on every plan: Claude Sonnet 4.6, Claude Haiku 4.5, GPT-4o, GPT-4o mini, Gemini 2.5 Flash, Grok 4.3 and DeepSeek V4 Flash.
• Frontier models need Individual or higher, or the 7-day trial: Claude Opus 4.8, GPT-4.1, o3, o4-mini, Gemini 2.5 Pro, Grok 4.20 (reasoning) and DeepSeek V4 Pro.

These limits apply only when you pay with Codira credits. With your own key for a provider, you can use any of that provider's models. For what each plan includes, see Plans and pricing.

Your request still goes through. Codira quietly swaps in the best model your plan allows, for example Sonnet in place of Opus. Settings keeps your saved choice and shows it with a lock badge.

Chat picks automatically between Claude and GPT, using your default model for each. It uses Claude for long conversations and for questions about planning, review or explanation. It uses GPT for "write", "fix", "add" or "build" wording. Plain chat never uses Gemini, Grok or DeepSeek.

Every Claude model, GPT-4o, GPT-4o mini, GPT-4.1, o4-mini, both Grok models and both Gemini models. o3 and the DeepSeek models read text only. If you attach an image while a text-only model is selected, the Composer switches to the closest model from the same provider that can read images.

Always OpenAI's GPT-4o mini. You can't change it. It works with your own OpenAI key or with credits. See Chat, Composer, completions and commands.

With your own key for that provider, yes. Codira passes the model name straight to the provider without checking it. With credits, Codira accepts only models on its approved list and refuses anything else.

Codira replaces retired model names, such as the Claude 3.x models, with today's default, so old settings and old project files keep working.

No. When you're signed in, requests for any provider you haven't added a key for go through the Codira gateway and are paid for from your credits.

A service on codira.com that calls the AI providers on your behalf with Codira's own provider accounts. It checks your sign-in instead of an API key and charges your credit wallet. It covers all five providers. The gateway passes your request through to the provider and doesn't save the content of your prompts or the replies. For each call it records the model, the token counts, the cost and a run ID, which is what you see in your usage history.

Yes. Paste them under Settings → API Keys, one per provider. The Current Routing card there shows how each provider is being reached. Calls to a provider with your key go straight to that provider, and you pay them directly. They don't use Codira credits.

Your own key always wins for that provider. Codira decides provider by provider, so you can use your own Anthropic key and pay for OpenAI with credits. You can change this at any time without restarting.

In the macOS Keychain, one entry per provider. They're never written to your project, never kept in plain text in Codira's settings file, and never included in crash reports. Older versions kept keys in the settings file; Codira moves any such key to the Keychain the next time it starts. Clearing a key in Settings deletes it from the Keychain.

@web searches send your search words to DuckDuckGo. The Design tool fetches the web page whose address you give it. Neither sends your code.

• Insufficient credits: top up, or add your own key for that provider.
• Rate limited: you've sent too many requests in a minute. The limit per minute is Hobby 60, Core and Core+ 200, Ultra 300, Teams 500, Enterprise 1,000 (see Plans and pricing).
• Model not allowed: the gateway doesn't accept that model. Pick one from the list.
• Model requires a higher plan: upgrade, or pick a standard model.

Brief network or provider hiccups are retried automatically after 0.6, 1.8 and 4 seconds. A reply that goes silent for 2 minutes is stopped.

Your sign-in has expired. Codira tries to renew it automatically. If that fails, it asks you to sign in again. You don't need to buy a key.

Providers you've added your own key for keep working. Providers you reach through the gateway pause until the service is back. Your code isn't affected either way.

They're the two ways the chat panel handles a request. You switch between them in Settings → AI Models → Execution mode.

• Freestyle: one fast agent acts on almost anything you type: a brief, a change request, or "yes" to carry on. Only real questions and small talk get a conversational reply.
• Orchestrate: only a clear instruction that starts with a build word ("add…", "fix…", "create…") goes to the full multi-agent team. Everything else is answered as chat.

Slash commands work the same way in both modes.

Freestyle.

A single agent writes the change and shows it as a patch card. When you click Apply, Codira builds the project, reads any real errors, fixes them and repeats until the build is clean, then runs your tests. To stay fast, Freestyle skips the Security and QA agents and the independent Verifier. When it's done, it offers Harden with Orchestrate to run those checks afterwards.

It runs the whole team in order: Planner → Implementer → Reviewer → Security → QA. After you apply the change, it checks the build, runs your tests, runs a UAT smoke check of the app, and finally asks the independent Verifier for a verdict. It's the most careful mode.

It takes your most recent Freestyle change and runs the checks Freestyle skipped: a security review, a fresh build and test run, and the Verifier. Click the Harden card after a Freestyle change, or type /harden (or /verify).

Ten:

Planner

Turns your request into ordered steps, with the files to change and what success looks like.

Implementer

Writes the change.

Reviewer

Checks the change against the plan and can send it back for revisions.

Security

Looks for leaked secrets, injection risks and unsafe data handling.

QA

Suggests tests.

UX Reviewer

Compares your interface with your design system, using a screenshot when your app is running.

Verifier

Reads the real build and test output and gives a plain-English verdict.

Auditor

Does a read-only bug hunt on existing code (/audit).

Explainer

Answers questions about your code (/explain).

Debugger

Explains a failed test or UAT check and offers fixes.

Yes, some of them. In the Agents panel (the sparkles icon), the Planner and Implementer are always on. The Reviewer, Security, QA and UX Reviewer can each be switched off.

Yes. While it works, the Implementer can open files, list folders, search your code and run safe commands, so it checks instead of guessing. Each file it opens is read up to 24,000 characters, a folder listing up to 300 entries, a search up to 40 results, and command output up to 8,000 characters. It can take up to 12 rounds of looking before it writes. Any command it runs goes through the command safety rules in Safety, review and undo.

Hobby

1

Core

4

Core+

6

Ultra

8

Teams · Team

6

Teams · Organization

8

Enterprise

Unlimited

Above your limit you see "Concurrent agent limit reached". For plan prices, see Plans and pricing.

A Full Review is a ⌘K Composer run with all five core agents (Plan, Implement, Review, Security, QA). Hobby includes 5 Full Review runs a month. Every paid plan has unlimited Full Reviews. At the Hobby limit, Codira suggests Fast mode instead.

Codira checks its work in this order, and each step runs only if the one before passed:

1. Install. If the change added packages, Codira asks you, then installs them with the package manager your lockfile uses.
2. Build. It runs your compile or type-check command, for example npx tsc --noEmit, npm run build or cargo check. Real errors go back to the Implementer, for up to 3 rounds of fixes.
3. Tests. It runs your test command if it finds one.
4. UAT smoke check. It loads your running app and looks for errors (see Context, memory, standards and testing).

Test and UAT failures share up to 2 rounds of fixes. A multi-step plan is checked again once at the end.

Builds get 300 seconds, agent commands 180 seconds, and dependency installs 600 seconds. If a build runs out of time, the change is reported as "applied UNVERIFIED", never as passed.

Yes. Settings → AI Models has "Verify build after apply" and the automatic UAT check. Both are on by default.

The Verifier reads what really changed on disk and the raw build and test output. It then gives one of four verdicts: Verified, Failed, Unverified (nothing ran that could prove it), or Suspected false green (a reported pass that the output contradicts). A fixed rule stops it from ever saying "Verified" if a command failed, the test suite didn't load, or nothing ran.

No. It means the change passed whatever your project's own build and tests could show. The verdict lists its limits.

Six automatic checks that run on every change before you see it. They don't use AI and cost no credits.

DEL

Code deleted that the request didn't call for

GROUND

References and imports to files or names that don't exist

REGEN

Whole files rewritten to change a little

STUB

Placeholder or unfinished code

PRESERVE

Exports that disappeared

UX

Breaks from your design system

Codira also checks that imports actually resolve. Most findings go to the Reviewer as advice. A critical unresolved import or critical stub turns an "approved" verdict into "needs changes".

A single recommendation above the diff that sums up every agent's verdict and every guard: Ready to apply, Apply with care or Don't apply yet, with the reasons. It's advice only. It never disables the Apply button.

Up to 2 rounds in the ⌘K Composer, and up to 3 rounds in a chat /plan.

Yes. A HIGH or CRITICAL security finding stops any automatic apply, in both the Composer and multi-step /plan runs. The run waits for you to review it.

Before planning, it reads your folder and file naming conventions so new paths match yours. After each step of a plan, it records what's really on disk, so the next step builds on the real state of your project.

Yes, by default. "Use current framework versions" looks up the latest stable versions of common frameworks on npm, crates.io and PyPI and gives them to the agents. Switch it off in Settings → AI Models.

Before commands like /explain, /ux, /fix and /understand, Codira warns you if you're low, but doesn't stop you. If the gateway refuses a request for lack of credits, you see a message with a button to top up (see Credits and usage). Work already applied stays applied.

No. The run stops and leaves the earlier steps in place. Each step has its own Time Machine checkpoint, so you can undo any of them (see Safety, review and undo).

After a test or UAT check fails, click Explain failure. The Debugger gives the root cause, a step-by-step account of what happened with file and line references, and 2 or 3 possible fixes. Choosing a fix asks the Implementer for a new change, which you review as usual.

The panel where you talk to the agents. Ask questions, describe changes, or use slash commands. How a plain message is handled depends on Freestyle or Orchestrate mode (see How Codira's AI works).

No. A build request without an open folder gets "Open a folder first". Questions are still answered.

Yes, in Freestyle. A reply such as "yes", "go ahead" or "do it" continues the agent's last proposal.

Yes, on your computer, one history per project. Temporary progress messages aren't kept, but important results are, such as the Verifier's verdict and run summaries.

Yes. Settings → AI Models → Chat message formatting: Formatted (the default) or Raw.

The Prompt Coach. Simple rules spot common gaps as you type, such as "Mention a file?" or "Narrow scope?". Click a chip to insert a template. It uses no AI and no credits. Click "hide" to turn it off for the session.

Press ⌘K in the editor and describe a change. The agents produce a diff, across one or more files, for you to approve. If you've selected text, the Composer works on that selection. Otherwise it works on the whole file. The agents may also edit outside your selection when the change needs it; the diff always shows everything they changed.

⌘↩

Run: Full Review with all five agents (Plan, Implement, Review, Security, QA)

⇧↩

Fast mode: Plan and Implement only, with no review and no retry

↩

New line

Esc

Close

On Windows, use Ctrl in place of ⌘.

Either you're not signed in on a plan that includes AI ("Sign in to use Composer"), or no file is open ("Open a file first"). Files over 500 lines show a cost warning but still open.

No. You see a diff for each file, the guard results and each agent's verdict. Reject throws everything away. Apply writes all the files together and saves a Time Machine checkpoint.

Each step shows its own diff and verdicts.

• Apply & continue: applies this step with its own checkpoint, then runs the next step against the updated files.
• Apply all (N): applies the remaining steps one after another, but stops at any step the Reviewer rejects, the security check blocks, or a critical guard flags.
• Stop: ends the run and keeps the steps already applied.

Yes. Paste or drop up to 4 images, in PNG, JPEG, WebP or GIF, up to 4 MB each. For example: "build this screen". Codira switches to a model that can read images if needed. There's no direct Figma connection, but you can paste a Figma frame as an image.

Yes. Grey suggestions appear as you type, 1 to 3 lines long, and Tab accepts them. It's off by default. Turn it on with the "AI Autocomplete" pill in the bottom panel's tab strip.

It sends up to 4,000 characters before your cursor and 1,000 after, to OpenAI's GPT-4o mini. It waits 400 ms after you stop typing, skips very short lines, and reuses answers for repeated requests. Each suggestion costs about $0.0001. There's a cap of 200,000 tokens a day, shown on the pill, which resets at midnight UTC.

/plan <goal>

Plans a feature step by step, then builds it after you approve (see Planning and building tools).

/fix <change>

A quick, targeted edit with the smallest possible change. No multi-step plan, and no Security or QA pass.

/migrate <change>

/plan set up for codebase-wide changes, such as framework upgrades or swapping a library.

/harden (or /verify)

Runs the checks Freestyle skipped on your last Freestyle change.

/audit <topic>

A read-only bug hunt. The Auditor reports issues, then the Verifier re-reads the code and marks each one CONFIRMED, PARTIAL or REFUTED. Nothing is changed.

/review <story>

Checks your uncommitted changes for a Project Plan story against its acceptance criteria. Advice only.

/explain

On its own: a 5-part tour of your project (stack, architecture, entry points, conventions, suggested questions). With a question: answers it. Never writes files.

/understand (or /comprehend)

Writes the project understanding file (see Context, memory, standards and testing).

/ux <focus>

A design review of your running app against your design system. Advice only.

/uat

Runs the UAT smoke check now (see Context, memory, standards and testing).

/brainstorm

Starts the Brainstorm interview (see Planning and building tools).

/genesis

Opens Genesis (see Planning and building tools).

/standards

Opens your team standards file, creating it if needed (see Context, memory, standards and testing).

/remember <fact>

Adds a dated line to your project memory (see Context, memory, standards and testing).

/scope <level>

Tells the Security agent how strict to be (see Context, memory, standards and testing).

/assets (or /catalog-assets)

Catalogues your project's images, logos and fonts for the agents (see Context, memory, standards and testing).

/reindex (or /index)

Rebuilds the code search index (see Context, memory, standards and testing).

/analyze

Re-runs the first-look project scan and shows it as a card. Uses credits.

/jira PROJ-123

Starts a plan from a Jira ticket. /jira on its own opens a ticket picker.

/launch

Opens the Launch form in the Ship panel. Nothing goes live until you press Launch.

You can also start a message with @web (below).

Use /plan for new features and changes across several files. Use /fix for a quick, targeted change from chat. Use ⌘K to edit the code you're looking at or have selected.

In Settings → Jira, enter your Jira Cloud site, your email and an API token, then click Test connection. The token is kept in the macOS Keychain. By default, Codira moves a ticket to "In Progress" when you pick it up and to "In Review" when the work is applied, with a comment listing the changed files. You can change both statuses or turn this off.

Yes. Start a message with @web <search words>. Codira searches DuckDuckGo (no API key needed), takes up to 10 results and gives them to the Planner. To search and act in one message, use @web <search words>; <what to do>. It only works at the very start of a message.

Name them in your message. Codira includes up to 5 files you name, up to 8 KB each. Your open tabs are included automatically. If a change refers to an image that doesn't exist yet, Codira offers a one-click import.

It writes the change to your files. Before writing, Codira saves a Time Machine checkpoint. All files in the change are written together, and each file is written safely, so a crash can't leave it half-written. See Safety, review and undo for what's applied automatically.

Yes, in three places:

• Settings → AI Models: your default model for each provider, and a model for each agent.
• The ⌘K Composer's Models picker: a one-off choice for a single run.
• Settings → This Project: shared models for everyone on the project.

Plain chat chooses between Claude and GPT for you (see AI models, providers and your own keys).

Type /plan and your goal. The Planner writes a step-by-step plan, shown as a card with Approve plan and Reject buttons. Nothing is written until you approve.

• A one-step plan produces a patch card that you apply yourself.
• A multi-step plan runs every step in order. Each step is applied automatically as soon as the Reviewer approves it, so the next step builds on it. The build, tests and UAT check then run once at the end.

Each step gets its own checkpoint, and a serious security finding stops the run for you to review.

If you'd rather approve every step yourself, use the ⌘K Composer, which asks at each step.

A way to start a new app from an idea. Describe your app in one sentence and AI writes a full brief, or paste a brief you already have. Genesis then:

1. Builds a first version of the site (Next.js projects only).
2. Has the Planner write an architecture brief (.codira/genesis.md), a backlog of epics and stories (.codira/backlog.json) and a build-ready brief (.codira/blueprint.md).
3. Lists prerequisites: what the planned stack needs (for example a database or Stripe), what you've already set up, and which stories each missing item blocks.

Open it with /genesis or from the activity bar.

A board view of the Genesis backlog, with four columns: Backlog → In Progress → Code Review → Ready for Release. Each story has a Run button that sends it through /plan. A finished run stops in Code Review for you to check; it never moves further on its own. A strip shows how much is done, and each run ends with a summary of what was built and what's next. The board's "Run review" button runs /review on that story.

A 10-question interview for when you're still shaping an idea. It covers what the app does, the audience, the data, how data flows, where users interact, team size, scale, compliance, integrations and priorities. Codira then writes .codira/blueprint.md, which every later /plan and /fix reads. It uses the Planner's model and makes a single AI request. Start it with /brainstorm.

Brainstorm if you're still working out the idea. Genesis if you already know what you want, as a sentence or a written brief, and want a working start plus a backlog.

A guided path for a brand-new project, in 7 steps: what are we building, set up the project, design system, your brief, build it, project plan, and get set up to build. You can leave and come back, because each step's progress is read from your files. The one-shot builder in the wizard makes Next.js projects only.

Several tools help:

• Create New Project runs the official setup tool for Next.js, Vite + React, Expo, SvelteKit, Astro or Remix. It doesn't use AI.
• The Setup Wizard and Genesis can build a first version of a Next.js site with AI.
• Architect's Stack tab installs the pieces you pick (below).
• `/plan` can build features into any project you open.

A visual planning tool, opened from the network icon. It has four tabs: Stack (technology tiles), Schema (a database diagram with 1:1, 1:N and M:N links), Tokens and Notes. It's saved to .codira/architect.json, and the Planner reads it on every run.

Each tile (framework, database, ORM, auth, payments, hosting and so on) maps to install commands from a fixed list of about 76 tiles. The commands come from that list, not from AI, and each one asks for your approval. Afterwards, AI can optionally add the config files. Commands that create a whole new project, such as create-next-app, are shown but never run.

Yes. It loads Prisma, Drizzle, SQLAlchemy and Django models automatically.

Codira watches your schema files (Prisma, Drizzle, Python models) and package files (package.json, Cargo.toml, requirements.txt, pyproject.toml). About 4 seconds after one changes, it compares them with the canvas. If there's something new, a message names the new tables or tools and offers Add to canvas. If you dismiss it, nothing changes, and Codira checks again after the next change.

Yes. Schema → Generate Files writes a Prisma schema, a Postgres SQL migration, a Drizzle schema, TypeScript types or Markdown docs. The file arrives as a normal patch card, with a checkpoint when you apply it. You can also copy the schema as Prisma, SQL or Markdown.

Only for Prisma, with Sync to source. Renaming a table is applied as a removal plus an addition, and you're warned first. For Drizzle, SQLAlchemy and Django, use Generate Files.

Yes. Architect's database assistant can only act through fixed tools: list, add or remove tables and fields, link tables, and build the database. It can't claim to have done something unless one of those tools did it. One click undoes its whole run.

Yes. Build database creates it on your running Postgres and only reports success for tables it has read back from the database.

Hobby: 1. Individual, Teams and Enterprise: unlimited. Sharing Architect documents with your organisation needs the organisation's owner to be on Teams or Enterprise. Shared documents can be viewed and edited by every member. If two people save at once, the second save is refused rather than overwriting the first.

A plain-English helper for founders, product managers and anyone who doesn't want to write prompts. Open it from the wand icon in the activity bar. It never writes code itself. It helps you decide what to ask for, then hands that to the agents in one click. It has three modes:

• Prompt: asks clarifying questions and drafts a precise /plan.
• Code Inspector: a read-only review that explains issues by severity, from low to critical, and drafts a /fix for each.
• Readiness: answers "are we ready to ship or scale?" with findings by area and a fix prompt for each gap. You can focus it on general, ship, scale, security or tests.

It chooses between Claude and GPT automatically, using your default model for each. It reads the same project context as the agents: your code, MEMORY.md and your standards.

Two tools help:

• Rename Symbol (⌘⇧R) renames a word everywhere in the project. It isn't AI and doesn't understand code: it matches whole words, case-sensitively, including inside strings, JSON and .env keys. You see a preview grouped by file, can untick files, and apply with ⌘↩. It has no undo (see Safety, review and undo).
• `/migrate` is for larger refactors that need AI, such as changing a pattern across many files.

A live preview of your running app, in the bottom panel. Click an element to see what it is and where it's defined. You can change its styles or text directly, or describe a change ("make this feel premium"), which opens the Composer for that element.

With precise targeting on, each element carries a source tag, so Codira edits exactly the element you clicked. This works with Next.js and Vite. For other setups, Codira looks for a unique text or class match and does nothing if it finds more than one. On Vite projects, Enable Canvas installs Codira's small inspector plugin as a development dependency, updates your Vite config and restarts the dev server.

• Quick edits: padding, margin, font size, border radius, font weight, text colour and background.
• Text and style changes typed into the editor.
• Structural edits: delete an element, or reorder it among its siblings.
• AI Visual Assist: Edit, Move, Duplicate or Insert, driven by a prompt. Images you attach are saved to public/.

Not when Codira can make them directly. If you change a colour, style or text and Codira can find the element in your code, it writes the exact value with no AI involved: no credits and no wait. Deleting an element is only ever done this direct way. If a direct edit isn't safe, Codira refuses rather than guessing.

Open it from the pen-nib icon and enter a website address. Codira fetches the page's real HTML and CSS, and AI turns them into design tokens (colours, type, corner radii) plus a design guide. You can then:

• Adopt into my tokens: update or add the values in your CSS theme.
• Re-skin: replace hard-coded colours across your project with the nearest colour from the new palette. A colour is either matched with certainty or left alone.
• Save `DESIGN.md`.
• Import the brand's logo, favicon and fonts into the Asset Hub.

• The UX Reviewer checks interface changes against your design system as part of a full review.
• `/ux` reviews a screenshot of your running app. Use /ux <focus> to point it at one area, for example "/ux the settings modal". Without a running app it reviews the code only.
• The UX guard flags breaks from your design system on every change.

Only files inside the project folder you opened. Every read, write, rename and delete is checked against that folder, and any path that leads outside it is refused. All AI file changes go through Codira's patch step, which saves a checkpoint first and writes each file safely.

Yes:

• It can't delete or rename anything inside .git.
• A change that deletes more than 5 files is refused unless you confirm it.
• A change that contains unresolved merge-conflict markers is refused.
• It can't write outside your project folder.

No. Commands are for looking, not editing. Codira compares your files before and after each agent command. If a command it judged safe changed anything, Codira puts things back (restores changed files and removes new ones) and tells the agent. Commands you approved yourself may change files, and Codira tells you what changed. Files you'd already changed before the command are never touched.

By default, single changes wait for you to click Apply, in both Freestyle and the Composer. These are the exceptions:

• Multi-step `/plan` runs apply each step automatically once the Reviewer approves it (see Planning and building tools).
• Composer "Apply all" applies the remaining steps until one is flagged.
• Verified checks (build, tests, UAT) run automatically after a change is applied, unless you turn them off.

A HIGH or CRITICAL security finding always stops automatic applying.

Each change shows a diff for every file, the patch guard results, each agent's verdict and the Apply Decision Pill. Read the diff, then Apply or Reject. Nothing in the diff is written until you apply it.

Codira's undo history for AI changes. Before each AI change is applied, Codira saves a snapshot of your whole project as a hidden git checkpoint. Open the clock icon to see each checkpoint's goal, the files it changed, and a diff for each file. Each project has its own timeline.

Open Time Machine (the clock icon), find the change and click Revert, or use Revert on the result card in chat. Files the AI changed go back to how they were, and files it created are removed. Ignored files such as .env aren't touched. Reverting keeps the checkpoint, so you can go forward again.

Not by itself. Checkpoints are taken just before AI changes are applied. A checkpoint does include any unsaved-to-git work you had at that moment, but edits you type between AI runs aren't checkpointed. Use git commits for those.

No. They're stored as hidden references, so they don't appear in git log, don't move your branch and aren't pushed.

Yes. The first time an AI change is applied, Codira sets up git in the folder and makes a starting commit of everything not ignored, called "Initial commit (Codira baseline)". Time Machine needs this.

Hobby

7 days

Individual (Core, Core+, Ultra)

30 days

Teams and Enterprise

Unlimited

Older checkpoints are removed quietly when you open a project, when your plan changes, and every 24 hours. For plan details, see Plans and pricing.

Yes. Builds, tests, package installs and the commands agents use to investigate all pass through one safety check. Each command is put in one of three groups:

Safe

What happens: Runs straight away

Examples: Builds, tests, linters, commands that only read

Needs approval

What happens: A native "Approve command" dialog asks you each time

Examples: npm install, pnpm add, yarn add, pip install, brew install, cargo install; git push, git reset --hard, git clean, git config; any rm, plus mv, rmdir, truncate, chmod, chown; find … -delete or -exec; one-line scripts such as node -e, python -c or bash -c; output written to an absolute or home path

Blocked

What happens: Never runs

Examples: Anything starting with sudo; rm -rf on /, your home folder or system folders; mkfs; dd if=; chmod -R 777 /; fork bombs; curl or wget piped into sh or bash; Windows format and recursive deletes of a drive or system folder

Approvals are never remembered; you're asked every time. The check runs in the app's core as well as its interface, so an interface bug can't skip it.

No. Codira splits commands on ;, &&, ||, | and new lines, checks every part, and applies the strictest result.

Agents are told not to. Their commands run in the background, can't be interactive, and are stopped after 180 seconds.

No. The built-in terminal is yours, like Terminal.app, and nothing you type there is checked or blocked. Only commands Codira runs for you go through the safety check.

• Harmful code the AI writes inside your project. Review the diff; Time Machine is your safety net.
• Install scripts in packages you approve. They run with your permissions.
• The terminal, and Ship actions you click yourself.

For chat and planning it sends:

• your open tabs, including up to 8 KB of the file you're looking at
• up to 5 files you name in your prompt, up to 8 KB each
• the 3 files that best match your request, up to 2 KB each, found with the code search index
• a folder tree, 3 levels deep and up to 80 entries
• for a new project, up to 6 starting files such as README and package.json, up to 2.5 KB each
• your project memory, team standards, project understanding, blueprint, Architect design and asset list

The Implementer may open more files while it works (see How Codira's AI works). Everything goes to the provider running that agent, directly with your own key or through the Codira gateway with credits.

Up to 5,000 files of up to 200 KB each. Only source, config and docs file types are indexed: JavaScript and TypeScript (including Vue, Svelte and Astro), Python, Rust, Go, Ruby, PHP, Java, Kotlin, Swift, C#, Scala, C and C++, CSS, SCSS, Less, HTML, JSON, YAML, TOML, XML, Markdown, shell, SQL, GraphQL, protobuf, Terraform, Lua and R. Hidden files and .env files are skipped. Building the index costs about $0.02 for a 1,000-file project. After that, re-indexing only covers changed files. Type /reindex to rebuild it.

Your saved OpenAI key is wrong or expired. Go to Settings → API Keys and clear the OpenAI key. If you're signed in, Codira falls back to the Codira gateway.

Yes, through Project Memory, a plain markdown file at .codira/MEMORY.md in your project. You can read, edit and commit it. It's added to every plan, implement and review request, and to chat.

Type /remember <fact> in chat, which adds it as a dated line. Or edit the file in the Memory section of the Agents panel. The .codira/ folder is only created the first time you save something.

It tells the Security agent how strict to be. The options are local-only, single-user, prototype, side-project, production, saas and library. Without it, Security assumes production-level caution, which can flag a lot on a personal project. Your choice is saved as a line in MEMORY.md.

Run /understand once. Codira reads your whole project and writes .codira/understanding.md, in two parts:

• Verified facts, read straight from your files with no AI: build and test commands, folder layout, how files import each other, and test fixtures.
• Observed patterns, suggested by AI and then checked against your code. Anything the code contradicts is dropped.

Every agent reads this file on every run. A banner offers a refresh when your project has changed a lot.

A catalogue of your project's images, logos and fonts, saved to .codira/assets.json. Every agent sees it, so they can refer to "the logo" and know how to use it. Add images with the explorer's Import image button: PNG, JPG, GIF, SVG, WebP, AVIF or ICO, copied into public/ (or src/assets/).

Three layers are added to the agents' instructions:

1. Codebase Baseline: learned automatically from your code and saved to .codira/baseline.json. It understands TypeScript and JavaScript, Python, Rust and Go.
2. Team Standards: your written house rules in .codira/standards.md.
3. Personal Style: five settings that apply only to you and stay on your computer: comment density, naming style, test depth, patch preference and how much the agent pushes back. Set them in Settings → Personal Style.

When the Baseline and Team Standards disagree, Team Standards win.

Type /standards, use Settings → Team Standards, or accept the banner Codira shows when a project has none. Codira creates .codira/standards.md from a starter template. Commit it so everyone shares it. Every agent reads it on every run, so edits take effect straight away. Only the first 2,000 characters or so reach the agents, so keep it focused. It works on every plan, and nothing is stored on Codira's servers.

A shared markdown document for your whole organisation, up to 20,000 characters, stored on codira.com. It's added to every teammate's agent requests, alongside each project's MEMORY.md. It appears in the Memory panel when you're in a team organisation. Any member can read it. Editing it needs the organisation's owner to be on Teams or Enterprise. If two people edit at once, you see "Team memory changed" instead of losing someone's work.

That's your choice. We suggest committing the shared knowledge files, such as MEMORY.md, standards.md, understanding.md, blueprint.md, genesis.md, architect.json and settings.json, so your team shares them. Machine-specific files (baseline.json, embeddings.json, chat transcripts and temporary files) are added to your .gitignore when you use the Ship panel. You can delete the folder at any time; Codira rebuilds what it needs.

A quick "does the app still load?" test after each applied change. Codira opens your running app in a hidden window, clears its stored state, loads the home page, waits for it to appear, takes a screenshot and collects console errors. The result is shown on a UAT card. It waits up to 8 seconds for the page to be ready and 60 seconds in total.

Your dev server must be running; if it isn't, the check is skipped. It's for web apps only. Codira injects its test script through a small local proxy, so it works with Next.js, Vite and similar setups without changes to your project.

Yes, in Settings → AI Models. It's on by default.

It runs the same smoke check whenever you ask.

Yes, automatically after AI changes are applied. It finds your test command from:

• the test script in package.json
• cargo test
• go test ./...
• Maven (mvn test or ./mvnw)
• Gradle (./gradlew)
• pytest in your Python virtual environment

It also finds these in subfolders of a monorepo. There's no separate "Run tests" button; run them yourself in the terminal at any time.

The QA agent suggests test cases as part of a review. Automatic test writing only produces vitest tests, for JavaScript and TypeScript projects.

Codira sends the real failure output to the Implementer for up to 2 rounds of fixes. If it still fails, the run stops and tells you. Click Explain failure for the Debugger's root cause and 2 or 3 possible fixes (see How Codira's AI works), or undo the change in Time Machine.

The agents can work in any text-based language. How much Codira can check its own work depends on the language:

TypeScript and JavaScript

Full checks: build, tests, import checks, all guards. Best on Next.js.

Python

ruff, pyright and a compile check, pytest, and import checks

Rust

cargo check and cargo test

Go

go build and go test

Java and Kotlin

Maven or Gradle compile and test

Anything else

No build or test check

• Go and Java have no import or name checks, and critical findings don't block automatically.
• Automatic test writing only produces vitest tests.
• The deletion and stub guards use simpler checks for Go, Java and Python.
• The build check only finds Go and Java projects at the top of your project folder, not in subfolders.

Next.js, SvelteKit, Astro, Vue CLI, Create React App, Vite and Expo (web). Codira looks in your project folder first, then in folders such as frontend/, web/, app/ and client/, and one level into apps/* and packages/*.

The rocket icon in the desktop app. It is where your work leaves your computer. It has three parts:

• Repo: set up git, commit and push, sync, switch or create branches, create a GitHub repository and open a pull request.
• Deploy: put your app online, either on Codira Hosting or with another provider.
• Tools: install or sign in to the GitHub CLI (gh) and the Vercel CLI.

There are three ways:

1. Launch on Codira puts the app on Codira Hosting at a name.codira.space address. See Codira Hosting.
2. Direct Vercel deploy runs the Vercel CLI for you and shows the .vercel.app link when it finishes.
3. Deploy links open the "import this repository" page at Vercel, Netlify, Render, Railway or Cloudflare Pages in your browser. You finish setup there once. After that, every push to GitHub deploys.

Vercel is the default suggestion, and it is the only outside provider Codira can deploy to directly. If Codira Hosting isn't available to you, typing /launch points you to Vercel or another host.

The Ship panel runs vercel --yes for a preview deployment, or adds --prod for production, and shows you the address. Before the first deploy, run vercel login once. If the Vercel CLI isn't installed, the panel can install it: the command runs in the visible terminal, so you can watch it and stop it with Ctrl+C.

No. For these, Codira only opens the provider's import page. The provider does the build and hosting under your own account with them. Vercel, Netlify and Render fill in your repository for you. On Railway and Cloudflare you pick it from a list.

• An account with that provider.
• Your project in a GitHub repository. The Ship panel can create one for you (private or public, under your account or any organisation you belong to) and push your code.
• The GitHub CLI (gh) installed and signed in. The panel shows "Connected as <your login>" when it is.
• For the direct Vercel deploy, the Vercel CLI installed and signed in.

The one your Mac already uses. Push and pull run your normal git, with your usual credential helper or SSH key. Creating repositories and pull requests uses gh. If more than one GitHub account is signed in to gh, the panel warns you. Codira has no "switch account" button, because switching in gh changes the account for your whole computer. Run gh auth login in the terminal instead.

Codira reads your PATH from a login shell, which loads .zprofile but not .zshrc. If your Homebrew or nvm setup is only in .zshrc, move the brew shellenv line to .zprofile and restart Codira.

Push and pull work with any git remote. Creating repositories, opening pull requests and the deploy links are GitHub only.

No. These aren't deploy targets.

It depends on where you deploy:

• Codira Hosting: at launch you tick which keys from your .env files the live app gets. Only those values are sent, encrypted. See Codira Hosting.
• Vercel, Netlify, Render, Railway, Cloudflare: the Ship panel has no step that sends your variables to these providers. Add them in the provider's own project settings.

Settings → Env Vars in the desktop app. Add KEY=value rows. The values are kept in your Mac's Keychain and hidden until you click Reveal. Each time you save, Codira writes them into a marked block in .env.local, keeps any lines you wrote yourself, and adds .env.local to .gitignore. Vite, Next.js, Astro, Remix and other tools that read .env files pick them up automatically. Because .env.local is never pushed, outside providers won't see these values unless you add them there.

Not yet. Variables are added one row at a time, and each project has a single set.

Not from Env Vars, because .env.local is added to .gitignore. Codira also adds its own internal files (such as .codira/embeddings.json and chat transcripts) to .gitignore. Anything else you commit is up to you. Codira Hosting checks your files for keys written straight into your code before it uploads them.

Yes. A short message such as /launch, "deploy this", "launch my app" or "publish this site" opens the Launch form. It never puts anything online by itself. You still press Launch.

It puts an app you built in Codira online at its own address, name.codira.space. You press Launch in the Ship panel. Codira packs your project, uploads it, builds it and runs it. The app stays online when your computer is off, and you don't need a separate hosting account.

Yes, for paid customers.

Anyone with an active paid plan. The free 7-day trial and Hobby don't qualify, because neither needs a card. This is a safeguard against phishing sites.

Anything that starts a web server: Next.js, Express, Vite, Python, Go and similar. Your server must listen on the port given in the PORT setting, which most frameworks do by default. Each app runs as one web service, with one build command and one start command.

No. There is no managed database. If your app needs one, use a database hosted elsewhere and give the app its connection details as a secret setting.

All apps run in New York, on DigitalOcean's App Platform. You can't choose a region.

1. Open the project in Codira.
2. Open the Ship panel. Under Launch on Codira, press Launch…, or type /launch in chat.
3. Choose an address and a size, and tick the secret settings the app needs.
4. Press Launch.

Progress shows as Uploading → Building → Starting → Live. The first build usually takes a few minutes. After that, the Ship panel shows the address with Open and Copy buttons, and the main button becomes Update.

By uploading. The desktop app packs your project and uploads it straight to Codira. If the project is on GitHub and you're signed in, the Ship panel first saves your changes there, so what's live matches a saved copy. That step is optional: choose Launch without saving to GitHub to skip it.

Every file git would keep. Anything in your .gitignore stays out. These are always skipped: .git, node_modules, .next, .vercel, .turbo, .cache, __pycache__, .venv, venv, .codira, symbolic links and .DS_Store files. .env and .env.* files are never uploaded, except .env.example, .env.sample and .env.template.

Yes: 100 MB compressed, 300 MB unpacked, and at most 10,000 files. The upload must finish within 5 minutes. Over the limit you see "The project is too large to launch." or "The project has more than 10000 files."

No. Press Update in the Ship panel. The current version keeps running while the new one builds, and visitors are switched over once it's ready. A failed update never takes down the live version. Each update is a full snapshot, so files you deleted from the project are removed from the app too.

Not launched yet, Waiting for review, Building, Starting, Live, Last launch failed, Paused: out of credits, and Suspended. Deleted apps disappear from your list.

Press Show details to read the build log. Or press Fix it with AI: Codira sends the end of the log to the agent as a /fix request, using your normal credits. Then press Update. The usual causes are a missing build or start command, or a server that doesn't use PORT.

Yes. Every launch, update, restore and settings change is saved as a version. Press Restore next to any earlier version that went live: the Ship panel shows the last 5 and the dashboard the last 20. The last 50 versions can be restored. A restore builds a new version from the old files. It changes code only; your secret settings stay as they are now. You can't restore while the app is paused, suspended or in review.

With secret settings. At launch, Codira reads your project's .env, .env.production, .env.local and .env.production.local files (later files win) and lets you tick which keys the live app gets. Only the ticked values are sent, encrypted, and stored by the hosting provider. Codira keeps only the names, never the values. Secrets are available both while the app builds and while it runs. If Codira finds a key written straight into your code, it warns you before launch.

A letter or underscore first, then letters, numbers or underscores, up to 128 characters. For example STRIPE_KEY or DATABASE_URL.

On the app's dashboard page, under Secret settings, add a name and value or remove one. In the Ship panel, Secret settings sends new values with your next Update. Removing a secret is done on the dashboard. The app redeploys to pick up the change, which appears as a new "Settings change" version.

A free <name>.codira.space address. Codira suggests one from your project name and checks straight away whether it's free. codira.space is kept separate from codira.com on purpose, so one bad app can't affect Codira's own site.

3 to 40 characters: lowercase letters, numbers and hyphens. It can't start or end with a hyphen, or have two hyphens in a row. Reserved names are refused, including www, api, admin, mail, support, billing, login, status, docs, test, staging, dev, team, legal and privacy, and anything starting codira- or ending -codira. Each address belongs to one app.

No. To use a different address, launch a new app and delete the old one.

Yes, up to 5 per app, such as www.mysite.com or mysite.com. Add it under Custom domains in the Ship panel or on the app's dashboard page. The app must already be launched and running (not paused, suspended or in review). Wildcard domains aren't supported, and each domain can be connected to one app only.

• A subdomain (www.mysite.com): one CNAME record, pointing at the address the panel shows you.
• A bare domain (mysite.com): four records at @:
• A 162.159.140.98
• A 172.66.0.96
• AAAA 2606:4700:7::60
• AAAA 2a06:98c1:58::60

The status reads Waiting for DNS, Active, or Check the records. HTTPS is set up automatically once the records work, usually in under an hour and at most a day. You add the records at your registrar yourself; there is no one-click setup.

Prices are in credits (1 credit = $0.01), shown for a full month of 730 hours:

Small

Memory: 512 MB

Credits a month: 800

About $ a month: $8

About credits an hour: 1.10

Medium

Memory: 1 GB

Credits a month: 1,600

About $ a month: $16

About credits an hour: 2.19

Large

Memory: 2 GB

Credits a month: 4,000

About $ a month: $40

About credits an hour: 5.48

X-Large

Memory: 4 GB, 2 CPUs

Credits a month: 8,000

About $ a month: $80

About credits an hour: 10.96

Yes. Every paid plan includes one Small app at no charge. If you make that app bigger, you pay only the difference above Small: an included Medium costs 1,600 − 800 = 800 credits a month. The included app is your oldest app that isn't suspended.

By the hour, from the same credit wallet you use for AI. There is no separate bill. Charges build up hourly and appear in your credit history once a day at 00:00 UTC (or sooner, once they reach 100 credits), as "Hosting: name.codira.space (size)". You pay only for hours an app is running. Apps that are not launched yet, paused, suspended or deleted cost nothing.

Yes, while it's running, from the Ship panel or the dashboard. The app redeploys on the new size and the new rate applies from the next hour. You can't resize an app that is paused, suspended or in review. Moving to a size that costs credits needs a positive balance.

Yes. The app's dashboard page shows credits used so far this month, a forecast to the end of the month (UTC), a daily breakdown for the last 7 days, and the approximate daily cost at the current size.

1. When your hosting will use up your credits within 3 days, Codira emails you (at most once a week).
2. When you reach zero, you get an email. Your apps keep running.
3. After 48 hours you get a "pauses tomorrow" email.
4. At 72 hours, apps that cost credits are paused: taken offline, but not deleted.

Your included app is never paused for lack of credits. This grace period is for hosting only; AI features stop at zero as usual.

Add credits. Paused apps restart on the next hourly run, so within about an hour.

On the app's dashboard page, press Delete… and type the app's address to confirm. The app goes offline for good, stops costing anything, its custom domains are disconnected and its address is freed. This can't be undone.

Yes. The app's dashboard page shows the build log and the live (run) log for the latest version, the last 64 KB of each. In the Ship panel, Show details shows the build log.

Yes. Restart on the dashboard restarts the current build without rebuilding, and the site stays up.

Partly. The dashboard shows data sent per day, CPU and memory use (average and peak) and restarts, over 24 hours, 7 days or 30 days. It doesn't show visitor or request counts. Today's data-sent figure appears tomorrow. If CPU or memory peaks at 85% or more, the dashboard suggests a bigger size.

Yes. Every live app is checked every 5 minutes at its codira.space address. After two failed checks in a row (an error, or no answer within 10 seconds), you get one "is down" email, and one "back up" email when it recovers. Custom domains aren't checked.

No. Hosting has no uptime guarantee and no credits for downtime. Keep your own backups.

No. Codira adds no analytics, cookies or tracking to hosted apps.

Every launch is checked automatically, and anything that looks risky is held for a person before it goes live. It is never rejected automatically. Launches are held when the address looks like a well-known brand, a page shows a password field next to a brand name, the code sends passwords or card details to a chat bot, or the project links to a known phishing or malware site. Reviews usually happen within a few hours during the day, or by the next morning overnight. While an update waits, the version already live keeps running. Size changes, restores, secret changes and new domains wait until the review is done.

It is taken offline, but its files, domains and secrets are kept while Codira reviews it. You can't launch, resize or restore it. The Ship panel shows "This app has been suspended. Contact support@codira.com." Suspended apps aren't billed and aren't deleted automatically.

Up to 25 apps at once (your own plus those of teams you own), up to 5 addresses reserved but not launched, and up to 10 new apps in any 24 hours. These limits are the same on every paid plan.

Yes. If you belong to a team workspace, the launch form asks whether the app is for you or the team. A team app is billed to the team's owner. Every member can see it, update it, restart it, restore versions and read logs and usage. Owners and admins can also change its size, secrets and domains, and delete it.

Yes, at codira.com/dashboard/hosting from any browser, or with Manage in the Ship panel. Everything except launching new code can be done there.

Use the form at codira.com/report or email abuse@codira.space. Choose phishing or fake login, malware, scam or fraud, copyright or trademark, spam, or other. Your email is optional. A person reads every report. Codira aims to act on phishing and malware within 4 hours during the day, or by the next morning overnight.

All prices are in US dollars.

Hobby

Monthly: Free

Annual: —

Credits each month: 200

Agents at once: 1

Individual Core

Monthly: $20

Annual: $220

Credits each month: 4,000

Agents at once: 4

Individual Core+

Monthly: $60

Annual: $660

Credits each month: 14,000

Agents at once: 6

Individual Ultra

Monthly: $200

Annual: $2,200

Credits each month: 50,000

Agents at once: 8

Teams · Team

Monthly: $40 per seat

Annual: $440 per seat

Credits each month: 8,000 per seat

Agents at once: 6

Teams · Organization

Monthly: $80 per seat

Annual: $880 per seat

Credits each month: 20,000 per seat

Agents at once: 8

Enterprise

Monthly: Custom

Annual: Custom

Credits each month: Arranged with sales

Agents at once: Unlimited

• No card needed.
• 200 credits a month.
• Base models only.
• 1 agent at a time.
• 5 Full Review runs a month.
• Time Machine history for 7 days.

Yes. Hobby has no time limit.

Everything in Hobby, plus:

• Frontier models (see below).
• Unlimited Full Review runs.
• Automatic credit top-up.
• Time Machine history for 30 days.
• Codira Hosting, with one Small app included (see Codira Hosting (codira.space)).
• Email support.

Usage only. The features are the same.

Price

Core: $20 a month

Core+: $60 a month

Ultra: $200 a month

Credits

Core: 4,000

Core+: 14,000

Ultra: 50,000

Agents at once

Core: 4

Core+: 6

Ultra: 8

Requests per minute

Core: 200

Core+: 200

Ultra: 300

"Frontier" models need Individual or above, including the trial: Claude Opus 4, GPT-4.1, o3, o4-mini, DeepSeek V4 Pro, Grok 4.20 reasoning and Gemini 2.5 Pro. Every other model Codira offers works on Hobby.

A Composer run that goes through the whole review process rather than fast mode. Hobby gets 5 a month. Every paid plan is unlimited. At the limit, Codira suggests fast mode instead.

Everything in Individual, plus:

• One bill and admin for the whole team.
• Shared team memory and shared Architect.
• Single sign-on (SSO).
• Audit logs.
• Time Machine history kept with no time limit.
• Priority support.
• Up to 500 requests a minute.

Team is $40 a seat a month with 8,000 credits a seat and 6 agents at once. Organization is $80 a seat a month with 20,000 credits a seat and 8 agents at once.

As many as you have paid seats, and the owner counts as one. You can't invite someone beyond that until you add a seat.

A custom plan arranged with Codira's sales team; it isn't sold through the website. It never expires, has unlimited agents at once, up to 1,000 requests a minute, a monthly spending cap of $10,000, and credits arranged with sales. Pricing is usually a yearly commitment, aimed at 50 or more seats or teams with special compliance needs. Contact sales@codira.com or use the contact-sales page; a person replies within one business day.

Yes. Paying yearly costs 11 months' price for 12 months, about 8% off. For example, Core is $220 a year. This applies to Individual and Teams plans.

Yes, per seat:

1–9

None

10–24

10%

25–49

15%

50 or more

20%

You can buy up to 1,000 seats in one checkout. The discount is set when you buy; it isn't recalculated if you change seats later.

Yes. Every new account gets 7 days of Individual Core, with no card, plus a one-time 1,000 credits. It starts when you confirm your email. There is one trial per account and per real inbox: dots and "+tags" in Gmail addresses don't count as new inboxes. If you have ever had a subscription, including an earlier trial, you can't start another. The trial is sometimes called the "Pro trial"; it is the same thing.

You move to Hobby automatically and are never charged, because the trial isn't a paid subscription. Any of your 1,000 trial credits you haven't used stay in your wallet. Your projects and history stay too. The trial doesn't get monthly credit drops, and it doesn't include Codira Hosting.

Not for Hobby or the trial. Individual and Teams plans need a card. Enterprise is arranged with sales.

Per minute: Hobby 60, Core and Core+ 200, Ultra 300, Teams 500, Enterprise 1,000. Over the limit, the request is refused with a "rate limited" message and you can try again shortly.

Hobby 7 days, Individual 30 days, Teams and Enterprise with no time limit. Older checkpoints are cleared automatically.

Yes. Old plans keep billing at their old price and keep working. Solo and Pro count as Core, and Team counts as Team. They are no longer sold.

A unit of usage worth $0.01. Plans include credits each month, and you can buy more.

At 4 times what the AI provider charges Codira, turned into credits at 100 credits a dollar and rounded up. Each call costs at least 1 credit. Input the provider reads from its cache costs 10% of the normal input price, and writing to the cache costs 125%.

For a request with 10,000 tokens in and 1,000 tokens out, it is about 18 credits on Claude Sonnet 4, 90 on Claude Opus 4, 6 on Claude Haiku 4.5, 14 on GPT-4o, 1 on GPT-4o-mini, 3 on Gemini 2.5 Flash and 1 on DeepSeek V4 Flash.

• AI in the desktop app when it goes through Codira (chat, planning, building, review, security, testing, completions and the code-search index).
• The AI features in the website's project tools. These show as "ai_pm" in your history.
• Codira Hosting, charged by the hour (see Codira Hosting (codira.space)).

No. If you've added your own key for a provider in Settings → API Keys, calls to that provider go straight to it and you pay the provider directly. Codira decides provider by provider, so you can use your own Anthropic key and pay for OpenAI calls with credits. Your key always wins for its provider.

Before each call, Codira checks that your balance could pay for one small request on the chosen model. That is about 8 credits for Claude Sonnet and 36 for Claude Opus. Below that, the call is refused and the app offers a button to top up.

Yes. The real charge is taken after the AI answers, so a large request can take you below zero. The app then shows a negative-balance warning. New calls are refused until you add credits.

You are charged for what the provider reports it used. If no usage report arrives within 5 minutes, that call isn't charged.

Once per calendar month (UTC), not on your billing date. They usually land the first time the desktop app checks your account that month. On Teams, the amount is the per-seat credits times your seats.

Yes, as one-time packs from Dashboard → Credits:

Starter

Price: $20

Credits: 2,000

Cost per credit: 1.0¢

Growth

Price: $50

Credits: 5,500 (+10%)

Cost per credit: 0.91¢

Scale

Price: $100

Credits: 12,000 (+20%)

Cost per credit: 0.83¢

Power

Price: $250

Credits: 32,500 (+30%)

Cost per credit: 0.77¢

Discount codes aren't accepted.

On Dashboard → Credits, click Buy now and pay in Stripe Checkout. In the desktop app, Settings → Account & Subscription → Credits → Buy credits opens the same page in your browser. When you come back, the page shows "+N credits added to your wallet". If you cancel, it shows "Checkout canceled. Your wallet was not charged."

Click the ⟳ refresh button in Settings, or quit and reopen Codira.

When your balance falls below a level you choose, Codira charges your saved card for a pack you choose. It is off by default and needs a paid plan; on Hobby you see "Auto-topup is available on paid plans." Settings:

• Threshold: 0 to 100,000 credits (default 500).
• Pack: Starter, Growth, Scale or Power (default Starter).
• Daily limit: 1 to 10 top-ups (default 3).

Top-ups are at least 10 minutes apart, need a default card, and respect your monthly spending cap.

Usually because the daily limit was reached (it pauses and shows a banner), the spending cap would have been passed, there's no default card, or the card was declined. The last error is shown on the Credits page. Turning auto top-up back on clears a daily-limit pause.

A limit on what Codira charges your card in a calendar month: subscriptions, credit packs and automatic top-ups together. Choose $50, $100, $200 (the default), $500, $1,000, $2,500 or $5,000 on the Credits page. Support can set it higher, and Enterprise is $10,000. A meter shows your spending, turning amber at 80% and red at 100%. Once you reach the cap, new packs, top-ups and upgrades are refused until the 1st of next month.

Raise the cap on the Credits page and try again. The total can take up to a minute to update.

• Dashboard → Credits: your balance, lifetime purchased, lifetime used and your last 25 transactions.
• Desktop app, Settings → Account & Subscription → Credits: your balance (live), lifetime totals and recent transactions, with your full history available.
• Hosting: each app's cost on its dashboard page.
• Teams: the workspace's Usage tab lists each member's credits used, runs and Full Reviews this month (UTC).

Each line shows the model, the amount of text processed and the credit cost.

In the desktop app, yes: it warns you when your balance falls under 500 credits and when it goes negative. For hosting, Codira emails you before and after you run out (see Codira Hosting (codira.space)).

Stripe. Codira never sees or stores your card number.

Cards only. Prices are in US dollars. Enterprise terms are arranged with sales.

Pick a plan on the pricing page. You pay by card in a Stripe form on the page, and the plan starts once the first payment succeeds. Teams plans let you choose 1 to 1,000 seats at checkout.

Yes, monthly or yearly, until you cancel.

From your workspace's settings, Billing tab, Open billing portal, or from the desktop app's account settings. It opens your own Stripe portal, where you can change your plan's billing, seats and card, and see past invoices.

In the Stripe billing portal. Codira picks up the new number automatically. Lowering seats below your number of members doesn't remove anyone; it only stops new people joining.

On Dashboard → Subscription, under Payment method, click Update (or Add card). The new card becomes the default for your subscription and for automatic top-ups. You can also change it in the Stripe billing portal.

On Dashboard → Subscription, under Payment history: your last 24 Stripe invoices, with date, number, amount, status and a PDF. Older ones are in the Stripe billing portal.

Stripe retries the card, and your subscription shows the status Stripe gives it, such as past due. Under the Terms, paid access may be suspended if you don't pay. A failed automatic top-up is shown on the Credits page.

Dashboard → Subscription → Cancel, then confirm "Cancel at end of billing period?" You keep your paid features until the period ends, and the button changes to Resume plan so you can undo it before then. Self-serve cancellation never ends a plan early. You can also cancel by contacting support.

They stay in your wallet. When the period ends you move to Hobby, so frontier models and automatic top-up switch off.

Sign in at codira.com and confirm your email. If you're signed out you go to the sign-in page; if your email isn't confirmed you're asked to confirm it first.

• Workspace: Overview, Projects, Hosting, Docs and Bug reports.
• Account: Subscription, Credits and Settings.

Your profile opens from your picture in the sidebar. The sign-out button is in the sidebar and the top bar.

A greeting, your plan and its status, how many days until it renews or ends (with the date), a Manage plan or See plans button, and quick links to Docs, Credits and Settings.

Codira's project and task tools for you and your workspace: projects, tasks, boards and sprints.

A list of your Codira Hosting apps. Open one to see its cost so far and forecast, traffic, versions, logs, size, custom domains and secret settings, and to restart or delete it. See Codira Hosting (codira.space).

Codira's user guide: installing, the agents, features, plans and credits, settings, shortcuts and troubleshooting.

Where you report a bug: a title, a description, a severity (low, medium, high or critical), and optionally steps, environment, a link and up to 8 screenshot links. You see only your own reports and can reply back and forth with support. The status moves through open, triaged, in progress, then fixed, won't fix or duplicate.

Your current plan and price, whether it renews or cancels and when, Cancel or Resume plan, plan choices, your card under Payment method, and Payment history with invoice PDFs. See Billing, subscription changes and refunds.

Your balance, lifetime credits bought and used, automatic top-up settings, your monthly spending cap, the credit packs and your last 25 transactions. See Credits and usage.

Your display name and your password. A new password must be at least 8 characters, and you need your current one if you have one. Your email address is shown but can't be changed here; contact support to change it.

Your picture (JPG, PNG, GIF or WebP, up to 2 MB, cropped to a circle), display name (up to 80 characters), bio (up to 280), your public @handle and the "Daily personal digest" email setting.

For team workspaces. The Billing tab shows your own subscription and opens your Stripe billing portal. The Usage tab shows each member's credits, runs and Full Reviews this month.

Not on the website. Your own AI provider keys go in the desktop app under Settings → API Keys, and are kept in your Mac's Keychain.

Not from the dashboard. Email support@codira.com, or privacy@codira.com for privacy requests.

• Verify your email: a 6-digit code when you sign up or ask to resend.
• Reset your Codira password: a link that works once, for 1 hour.
• Your Codira download: the installer link, if you sign up through the download page.
• Team invitation: when a workspace admin invites you.
• Hosting credits: running low, run out, pausing tomorrow, and paused.
• Hosting app down and back up.
• Daily personal digest: only if there's something new for you. Turn it off on your Profile page.

Look in your spam or promotions folder. Verification codes can be resent once a minute, and password-reset links up to 5 times in 15 minutes.

A workspace (also called an organisation) holds your members, sub-teams, web projects, shared Architect docs, Team Memory, audit log and single sign-on settings. Every account belongs to at least one workspace.

The first time you sign in, Codira creates a workspace called "<your first name>'s Workspace", with you as owner. Any older projects you own are moved into it. You can't delete your Personal Workspace.

Open the workspace switcher and choose Create workspace, then give it a name. You become its owner. Anyone can create one on any plan. Without a Teams subscription, though, the workspace has only one seat (yours), so you can't add anyone to it.

Use the workspace switcher. Codira remembers your choice. If you pick a workspace you're no longer a member of, it falls back to your default workspace.

Yes, if you're an owner or admin. Use the General tab under Settings → Organization. Names can be up to 80 characters, and each slug must be unique across Codira. Both changes are recorded in the audit log.

Owners and admins go to Settings → Organization → Members, enter an email address and choose a role (Member or Admin). Codira emails the person an invitation ("<name> invited you to <workspace> on Codira"). They don't need a Codira account first.

No. Each invitation is a single-use link sent by email to the person you invited. There's no "anyone with the link can join" option, and the link isn't shown in the members list.

14 days. If you invite the same email again, Codira sends a new link, restarts the 14 days, and the old link stops working.

No. A seat is used only when the invitation is accepted. If all your seats are full when someone tries to join, they see a message saying so. Their invitation stays valid, and the same link works once you add a seat.

Who invited them, to which workspace, and with which role. If they're signed out, they can sign up (with their email filled in) or log in, and then come back to click Join.

Yes. Owners and admins can revoke a pending invitation from the Members tab. The link stops working, and the revocation is recorded in the audit log.

Three: owner, admin and member.

Owner

Everything an admin can do. The owner can't be demoted or removed.

Admin

Rename the workspace and change its slug. Invite, revoke and remove members, and switch people between member and admin. Create, edit and delete sub-teams and manage who's in them. View the audit log. Set up SSO. Delete any shared Architect doc.

Member

See the member list and sub-teams. Read shared Architect docs and Team Memory. Create projects and work on every project in the workspace.

Groups inside a workspace, for example "Payments" or "Mobile". Owners and admins create them with a name, description and colour, and add people who are already in the workspace. Projects can be assigned to a sub-team. Deleting a sub-team detaches its projects but doesn't delete them.

That depends on the workspace owner's subscription:

• Teams (Team or Organization): as many active members as seats paid for. The owner counts as one.
• Enterprise: unlimited.
• Every other plan (Hobby, Individual, trial): 1 seat, the owner.

Team

Monthly: $40

Annual: $440

Credits per seat each month: 8,000

Agents at once: 6

Organization

Monthly: $80

Annual: $880

Credits per seat each month: 20,000

Agents at once: 8

Annual billing charges 11 months for 12. Prices are in US dollars. For full plan details, see Plans and pricing.

Yes, per seat, applied at checkout:

1–9

None

10–24

10%

25–49

15%

50 or more

20%

You pick between 1 and 1,000 seats at checkout. To change the number later, open the Stripe billing portal from Settings → Organization → Billing. Codira picks up the new number automatically.

Nobody is removed. New people just can't join until you're back under the limit.

The workspace owner, who pays.

Yes. The Usage tab lists each member's credits used, number of AI runs and Full Reviews for the current month (UTC), biggest user first.

Owners and admins can remove anyone except the owner from the Members tab. Removing someone also takes them out of every sub-team.

Projects they created stay in the workspace, and everyone can keep working on them.

Yes, for owners and admins on Teams and Enterprise, under Settings → Organization → Audit log. It records:

• members invited, joined, removed and changed between roles, and invitations revoked
• workspace renames and archives
• shared Architect docs created, updated and archived
• Team Memory updates
• SSO sign-ins

You can filter by action and page through entries. Teams keeps 90 days. Enterprise keeps everything.

Yes, on Teams and Enterprise. Owners and admins set it up per workspace under Settings → Organization → SSO, using either:

• OIDC: Google Workspace, Microsoft Entra ID or any OIDC provider.
• SAML 2.0: Okta, Microsoft Entra, OneLogin, ADFS or any SAML 2.0 provider.

• OIDC: paste your provider's discovery URL, client ID and client secret.
• SAML: give your identity provider the metadata URL Codira shows you (or the entity ID and ACS URL from it). Then paste in the IdP entity ID, the IdP SSO URL and the IdP's X.509 signing certificate.

You can also limit sign-in to up to 10 email domains, for example yourcompany.com.

Codira creates their account automatically, with no password, and adds them to the workspace as a member.

The controls Codira has in place: encryption at rest and in transit, authentication and audit logging.

One Markdown document per workspace, up to 20,000 characters. The desktop app adds it to every planner, implementer and reviewer prompt for every member, on top of each project's own .codira/MEMORY.md. Use it for things every teammate's agents should know, such as conventions, product terms or "never touch the billing module without review".

• Any member can read it. It appears in the Memory panel in the app when you're in a team workspace.
• Editing needs the workspace owner to be on Teams or Enterprise.
• If two people edit at once, the second save is rejected with "Team memory changed" so nobody's work is silently overwritten.

Architect diagrams (stack and schema) published to your workspace so every member can view and edit them. Publishing needs the workspace owner to be on Teams or Enterprise. Each doc has a version number, and if someone else saved first, your save is rejected so you can reload. The person who created a doc, or an admin, can archive it.

No. Team Standards live in .codira/standards.md in your repository, so they work on any plan and are shared through git.

A project-management tool in the web dashboard: projects, tasks, boards, sprints, a roadmap, reports and automations, plus some AI helpers. Find it under Dashboard → Projects.

No. Web projects, tasks and sprints live only on codira.com. The desktop app has its own, separate Project Plan board built from a Genesis backlog stored in your repo (.codira/backlog.json). The two aren't connected.

Every plan, including Hobby. There are no plan limits on projects or tasks. Only the AI features cost credits.

Go to Dashboard → Projects → New project. Give it a name and, if you like, a colour, a description and a sub-team. Codira makes a short key from the name (for example "PAY"), unique within the workspace, and numbers tasks PAY-1, PAY-2 and so on. You become the project owner.

Every member of the workspace can see and work on every project in it. There are no private projects. The project's member list is used for assigning tasks and @mentions. Projects from other workspaces are hidden while you're in this one.

The project owner adds them by email. The person must already have a Codira account and be in the workspace. No email is sent.

Delete the project, add or remove project members, manage custom statuses and manage automations.

It's deleted for good, with all its tasks, comments, activity, sprints, statuses, automations and notifications. There's no undo. Archive it instead if you might want it back.

• Types: task, bug, feature, epic, story, subtask, improvement, incident, AI task, research, QA and deployment.
• Priorities: low, medium, high and urgent.
• Default statuses: Backlog, To do, In progress, In review and Done.

Up to 20 tags and 50 dependencies per task, comments up to 5,000 characters, and story points from 0 to 1,000.

Yes. The project owner can add, rename, recolour, reorder and delete statuses in the project settings. Each status belongs to one of three groups: to do, in progress or done. A project must always keep at least one status and at least one "done" status. When you delete a status, you choose where its tasks move.

Board (drag and drop), List, Backlog, Sprints, Roadmap and Reports. There's also a command palette and a detail view for each task.

A sprint goes from planned to active to completed. A project can have only one active sprint at a time. When you complete a sprint, unfinished tasks go back to the backlog or into a planned sprint you choose. You can't delete an active sprint. Deleting any other sprint returns its tasks to the backlog.

Yes. Any task with a start or due date shows as a bar. You can zoom by day, week or month, group by epic, status or sprint, and drag bars to change dates. Sprint date ranges and dependency arrows are drawn too. Codira doesn't check dependencies for loops, so keep an eye on circular ones.

• Velocity for the last 20 completed sprints, with a 3-sprint moving average.
• Cycle time and lead time (average, median and 95th percentile, in hours) from the last 50 finished tasks.
• Weekly throughput for the last 8 weeks.
• How many tasks are in each status now.
• A burndown chart for each sprint.

Yes, inside Codira (the bell icon). You're notified when you're assigned a task, when a task you watch changes status, when someone @mentions you, when someone comments on a task you watch, and when an automation sends one. You automatically watch tasks you create, are assigned, comment on or are mentioned in. The bell refreshes every 30 seconds. These notifications aren't sent by email.

Yes, with rules for each project, which only the project owner can manage:

• Triggers: status changed, assignee changed or comment added.
• Actions: change status, assign someone, add a tag, add a comment or send a notification.
• Conditions: up to 10 per rule, on priority, type, status, tag, or the old or new status.

When one rule sets off another, the chain stops after 5 steps, so rules can't loop forever.

Yes. An open project page updates live: who else is viewing, tasks created, edited, moved or deleted, new comments, and changes to sprints, statuses and automations. Refresh the page if something looks out of date.

The AI Studio has three:

• Generate tasks from a requirements document or PRD (10 to 32,000 characters). You get up to 40 suggestions (12 by default) to review, then add the ones you want in one go.
• Triage a task: suggests priority, type, story points, tags, severity, a possible root cause and next steps.
• Release notes from a sprint or a date range, in Markdown, changelog or customer-facing style.

Yes. Each AI call is charged to your own credit wallet based on the tokens it actually uses, with a minimum of 1 credit. A typical call costs 1–10 credits. You need at least 1 credit available to start one, or you'll see "insufficient credits". Adding the suggested tasks you've reviewed is free. Calls appear in your credit history labelled "ai_pm".

GPT-4o mini (the default), GPT-4.1 mini, GPT-4.1 nano, GPT-4o, GPT-4.1, Claude 3.5 Haiku, Claude Haiku 4, Claude 3.5 Sonnet, Claude Sonnet 4 and Claude Sonnet 4.5. Larger models cost more credits per call.

That means the AI service behind it is temporarily unavailable. Try again later, or check codira.com/status.

• Jira Cloud: pull tickets into plans and update them as work lands.
• GitHub, through the GitHub CLI (`gh`): create repos, push and open pull requests.
• Deploy hand-offs: Vercel, Netlify, Render, Railway and Cloudflare Pages, plus direct deploys with the Vercel CLI.
• Codira hosting: launch to codira.space (see Codira Hosting (codira.space)).
• Web search: @web uses DuckDuckGo, with no API key needed.
• AI providers: Anthropic, OpenAI, Google Gemini, xAI Grok and DeepSeek, through Codira credits or your own keys.

In the app, open Settings → Jira and enter your site (for example acme.atlassian.net), your Atlassian email and a Jira API token. Click Test connection. The token is stored in the macOS Keychain. Codira talks to Jira directly from the app, not through Codira's servers.

In your Atlassian account settings, under Security → API tokens. Create one, copy it, and paste it into Settings → Jira.

• /jira PROJ-123 pulls that ticket into a plan.
• /jira on its own opens a picker with up to 25 of your open tickets.
• When you pick a ticket up, Codira moves it to "In Progress".
• When the work is applied, Codira posts a comment listing the changed files and moves the ticket to "In Review".

You can change both status names, or turn off write-back entirely so Codira only reads from Jira.

No. Only Jira Cloud, with an API token. There's no OAuth sign-in for Jira.

No, not today.

Through your own tools:

• Push, pull and fetch use your system git, with whatever credentials you already use (Keychain helper, SSH keys, gh auth).
• Creating repos and pull requests uses the GitHub CLI (gh).

Codira never asks for your GitHub password and doesn't store GitHub tokens itself.

Yes, from the Ship panel. Choose private or public, and your personal account or any organisation you belong to. Codira creates the repo and pushes your project.

Yes. The Ship panel pushes your branch and opens a pull request against the base branch you choose, with a title, a description and an optional draft flag. It then shows the PR's number, link and state. You need gh installed and signed in.

Whichever account gh and git are signed in with on your machine. The Ship panel shows "Connected as <login>" and warns you if gh has more than one account signed in. There's no switch-account button, because switching changes the account for your whole computer. Run gh auth login or gh auth switch in the terminal instead.

Partly. Push, pull and sync work with any git remote. Creating repos and pull requests is GitHub only.

From Ship → Deploy:

• Vercel, Netlify, Render, Railway and Cloudflare Pages: Codira opens the provider's "import from GitHub" page in your browser. You finish setup there once, and from then on every push deploys. Vercel, Netlify and Render fill in your repo for you; on Railway and Cloudflare you pick it from a list.
• Vercel CLI: a direct deploy that shows your .vercel.app address. Run vercel login first. The panel can install the Vercel CLI for you.
• Codira hosting: Launch on Codira (see Codira Hosting (codira.space)).

No. Docker is used only to run a local Postgres database for development.

Not through a Figma connection. You can paste a screenshot or exported Figma frame into the Composer (up to 4 images, 4 MB each), and Codira builds from the image.

No, and these aren't planned.

No. Codira's Terms and Privacy Policy require you to be at least 18, or the age of majority where you live if that's higher.

No. Codira runs on macOS 11 or later and on Windows 10 and 11 (64-bit). Linux and Chromebook support are on the roadmap, with no date.

At codira.com/docs. The same docs are in your dashboard under Docs. In the app, Help → Codira Documentation opens them in your browser, because the app doesn't bundle its own copy.

Yes:

• In the app: a feature tour appears the first time you open a project. Each new project after that offers "First time in this project? Take the tour". You can replay it any time from Settings → About Codira → Show welcome tour.
• On the website: codira.com/tour is a slide-by-slide product walkthrough.
• A first thing to try: open any file, press ⌘K, type "add a 2-line docstring at the top explaining what this file does", and run it. You'll see the agents plan, write and review the change, then show you a diff to apply or reject.

Yes. Create New Project builds a starter app (Next.js, Vite + React, Expo, SvelteKit, Astro or Remix). For a guided start, the Setup Wizard takes you from "What are we building?" to a working Next.js project with a plan. Brainstorm (/brainstorm) helps you think through an idea with about 10 questions.

Codira is gathering interest for local meetups with demos, workflow showcases, talks, networking and early product previews. Cities listed include San Francisco, New York, Austin, Miami, London, Berlin, Singapore, Tokyo, Toronto, Dubai and Sydney. Join the invite list at codira.com/meetups with your name, email and city. It's free, and you'll get a confirmation email.

Not yet. Workshops are coming soon. Use Get notified on codira.com/workshops, or email hello@codira.com. Enterprise plans include an onboarding workshop for your team.

Fill in the form at codira.com/demo with your name and work email. Company, phone and "What do you want to see?" are optional. You'll get a confirmation email, and a person replies within one business day to schedule a walkthrough. You can also email sales@codira.com.

Use codira.com/contact-sales: your name, work email, company, role, team size and what you're exploring. A person replies within one business day, and there's no automated email sequence. A typical path:

• Day 1: a reply.
• Week 1: a 30-minute scoping call.
• Week 2: security and compliance documents under NDA.
• Weeks 2–4: a pilot and the contract.

The sales form is for Teams and Enterprise. You don't need to talk to sales to buy an Individual plan.

Codira suggests a two-week pilot with one engineer on a real project. Track features shipped, review time per AI change, bugs in AI-written code, and how fast the engineer gets around unfamiliar code. Plan on a one- to two-week ramp-up.

Use the form at codira.com/press (name, email and outlet required), or email press@codira.com for media relations or media@codira.com for corporate communications. Replies usually come within 24 hours during US business hours. Put "urgent" in the subject line if you're on a deadline.

At codira.com/brand. The colours are ink #1e1d1c, orange #ee8f37, canvas #f4f4f4 and grey #666462. Keep the wordmark at least 96 px wide. Use the assets for editorial purposes only, without implying that Codira endorses you.

Email partners@codira.com.

Yes. There are seven engineering roles (AI Systems, macOS, Runtime Infrastructure, Frontend Platform, Security & Compliance, AI QA/UAT, and Infrastructure/DevOps) and five go-to-market roles (Enterprise Account Executive, Founding Sales Engineer, VP Enterprise Sales, Developer Relations, and Partnerships). Apply by emailing careers@codira.com; there's no application form. Details are at codira.com/careers.

Only what a feature you use needs:

Run an agent, chat, Composer, Sidekick and so on

What's sent: Your prompt and the code context it needs

Where to: The AI provider, directly (your own key) or through Codira's gateway (credits)

Build the semantic search index

What's sent: File contents, up to 7,500 characters per file

Where to: OpenAI's embeddings service, directly or through the gateway

Use autocomplete

What's sent: Up to 4,000 characters before the cursor and 1,000 after

Where to: OpenAI, directly or through the gateway

Use @web

What's sent: Your search words

Where to: DuckDuckGo

Launch on Codira hosting

What's sent: Your project files (never .env files)

Where to: Codira hosting

Push or pull

What's sent: Your commits

Where to: Your git remote

Use Jira

What's sent: Ticket updates and comments

Where to: Your Jira Cloud site

Sign in, check your plan, buy credits

What's sent: Account and billing details

Where to: codira.com and Stripe

The app crashes (unless you turn it off)

What's sent: An anonymous crash report

Where to: Sentry

Your files, chat history, Architect docs, memory files and Time Machine checkpoints stay on your computer.

For planning and chat, Codira sends:

• your open tabs, including up to 8 KB of the file you're working in
• up to 5 files you name in your prompt (8 KB each)
• the 3 files that best match your request (2 KB each)
• a folder tree, 3 levels deep and up to 80 entries
• for a new project, up to 6 starter files such as README and package.json
• your project memory, standards and design notes

The Implementer can also open more files itself while it works.

Not automatically. .env files aren't indexed or added to context. But if you name a .env file in a prompt, or an agent opens it while working, its contents go to the AI provider. Keep secrets in Settings → Env Vars and don't ask agents to read secret files.

When you use Codira credits, your requests pass through Codira's gateway to the AI provider. The gateway doesn't save the prompt or the response. For each call, it records only the model, the token counts, the cost and a run ID, for billing. Codira does store content you choose to put on codira.com: Team Memory, shared Architect docs, web projects and tasks, bug reports, and anything you launch on Codira hosting.

Yes, anonymously, and it's on by default. Reports go to Sentry and contain:

• the stack trace
• file paths cut down to the file name
• the app version and operating system
• a random anonymous ID stored on your computer

They never include your code, chat messages, API keys (key-like text is stripped) or console messages. Only crashes in the app's interface are reported.

Settings → About Codira → Send anonymous crash reports. The change takes effect the next time you open Codira.

In the macOS Keychain, never in plain text on disk and never in your repo:

• your own AI provider keys
• your Codira sign-in token
• your Jira API token
• each project's environment variables

macOS may ask once whether Codira can use the Keychain. Choose Always Allow. Older versions kept keys in a settings file; Codira moves them into the Keychain automatically.

Yes. Your sign-in token is kept in the Keychain and refreshed when Codira opens, so you don't have to go back to the browser each time. This means anyone using your unlocked computer can use Codira as you, so lock your screen when you step away.

Only the project folder you open. Every file read, write, rename and delete is checked, and anything outside the project is refused. The exceptions are files you pick yourself (such as an imported image) and creating a new project folder.

Only safe ones (builds, tests, linters, reads) run without asking. Codira sorts every command it runs for you into one of three groups:

• Safe: runs straight away.
• Needs your approval every time: installing packages, git push, git reset --hard, rm, mv, chmod, one-line scripts, and writes outside the project.
• Never runs: sudo, deleting system or home folders, disk formatting, fork bombs, and piping a download straight into a shell.

Chained commands are checked piece by piece. If a "safe" command changes your files anyway, Codira undoes the change and tells you. Commands you type into the terminal yourself aren't checked.

• Harmful code the AI writes inside your project. Review every diff, and use Time Machine to undo.
• Install scripts in packages you add, which run with your permissions.
• Commands you type into the terminal and Ship actions you click yourself.

No. It's signed and notarised by Apple and uses the Hardened Runtime, but not the App Sandbox. Codira confines itself to your project folder in its own code instead, as described above.

Each update is signed twice: Apple's notarisation covers the app, and a separate Codira key signs every update package. The app refuses any update whose signature doesn't match.

Yes. Traffic to codira.com is encrypted with TLS, and the database (MongoDB Atlas) is encrypted at rest. Card details are held by Stripe and never reach Codira.

In the United States. codira.com and its database run on Render in Oregon, with MongoDB Atlas. Apps on Codira hosting run on DigitalOcean in New York. Your code, chats and settings stay on your own computer unless a feature sends them, as described above.

Staff don't read customer code in normal operation, and production access is limited to a named on-call rotation using hardware security keys. The gateway doesn't store your prompts or code.

• Account details: name, email, a hashed password, and optional phone number, handle, bio and avatar.
• Your subscription, credit balance and credit history.
• Payment details, held by Stripe.
• Support messages and bug reports.
• Device, browser and IP information, logs, cookies and website analytics.

Codira doesn't sell personal data or use it for cross-site advertising.

Only as long as reasonably needed for the service, security, disputes and legal obligations. There are no fixed periods.

There's no self-serve option yet. Email privacy@codira.com or support@codira.com. You can also use your rights to access, correct, delete, move or restrict your data, or to withdraw consent, through privacy@codira.com. Codira may need to confirm your identity first. Invoices are always available in the Stripe billing portal.

Not for customer accounts yet. Teams and Enterprise workspaces can use single sign-on so that sign-in follows your identity provider's rules, including its MFA (see Teams and workspaces).

Yes, 18, or the age of majority where you live if that's higher.

Delaware law. Disputes go to binding arbitration, and jury trials and class actions are waived.

Email security@codira.com. Codira acknowledges reports within one business day, prefers coordinated disclosure, and can credit you in the changelog if you'd like.

A custom plan (see Plans and pricing) arranged with Codira's sales team, usually as a flat annual commitment. It's aimed at organisations with 50 or more seats or with special compliance needs. It isn't available through self-serve checkout.

• Unlimited members and unlimited agents running at once.
• Credits allocated to suit you instead of a fixed monthly amount.
• The highest rate limit: 1,000 requests a minute.
• A spending cap of up to $10,000 a month, or higher by agreement.
• Unlimited Time Machine history and audit-log history.
• SSO with OIDC or SAML 2.0.
• DPA and BAA on request.
• A custom contract, a dedicated customer success manager, a Slack Connect channel and an onboarding workshop.

Use codira.com/contact-sales or email sales@codira.com. A person replies within one business day. Codira sets up Enterprise on your account by hand, and it doesn't expire like a monthly subscription.

Yes. Enterprise uses a custom master services agreement, and data residency can be discussed in the contract. Today all of Codira's servers are in the United States (see Security and privacy).

After the scoping call, Codira can provide, under NDA:

• its security and compliance documents
• answers to your security questionnaire
• a DPA, and a BAA if you need one

Billing is in US dollars. Teams and Individual plans pay by card.

Codira opens codira.com in your default browser. Sign in there and click Authorize, then you'll see a page saying you can close the tab, and the app signs in. The app waits up to 5 minutes. If you have more than one Codira account, you'll be asked to pick one.

Find the 6-digit code Codira emailed you, enter it on the website, then click I've confirmed — retry. Use Resend to get a new code, once a minute. A new code replaces the old one.

• invalid_client or invalid_redirect_uri: your copy of the app isn't recognised. Update to the latest version from codira.com/download. If it continues, email support@codira.com.
• email_unverified: confirm your email first, as above.
• network_error: the page couldn't reach Codira. Check your connection and codira.com/status.
• missing_parameters: the page wasn't opened from the app. Start sign-in from the app again.

Your sign-in renews itself for up to 30 days. If two copies of Codira share one sign-in, for example after you copy a settings folder to another computer, Codira treats it as a possible leak and signs out every copy for safety. Sign in again on each computer.

Your sign-in has expired. Codira tries to renew it automatically. If that fails, sign in again. You don't need your own API key.

Codira trusts your last known plan for up to 7 days without a connection. Connect to the internet and it will check your plan again.

Use Forgot password? on the sign-in page. The link in the email works once and for 1 hour. You can ask for up to 5 links in 15 minutes, and each new link cancels the old ones.

Current versions are signed and notarised by Apple, so you shouldn't see this. Download a fresh copy from codira.com/download and drag it into Applications. If you still see the warning, right-click Codira.app, choose Open, then click Open, and email support@codira.com to tell us.

You opened it from the downloaded disk image. Let it copy itself to Applications and relaunch. Running from the disk image stops updates working and stops you ejecting the image.

Codira's Windows app needs Microsoft's WebView2 runtime. Windows 11 and current versions of Windows 10 already have it, and the installer downloads it if it's missing. No administrator rights are needed.

Sync services clash with build tools and create duplicate "file 2.ts" copies, even inside .git. Move the project to a folder that isn't synced, such as ~/Projects, then reopen it. Codira never moves or deletes anything itself, and you can dismiss the warning for each project.

Codira checks for updates when it opens and every 6 hours. When one is ready, a banner in the bottom-right corner offers Install or Later. To check by hand, go to Settings → About Codira → Check for updates.

Versions 1.6.3 to 1.6.6 were never offered updates, because of a release problem that has since been fixed. If you're on one of those versions, download the latest version from codira.com/download and drag it into Applications. Updates then work again.

Download the latest version from codira.com/download.

Not yet. Download the latest installer from codira.com/download and run it over your current version.

Your wallet doesn't have enough for the request. Buy a credit pack, turn on automatic top-up, or add your own API key for that provider. Before each request, Codira checks that you could pay for a small one on the model you've chosen: about 8 credits on Sonnet or 36 on Opus. So a balance of a few credits can still be refused.

Codira charges after the AI replies, so one large request can take your balance below zero. Add credits to continue.

Click the refresh button next to your balance, or quit and reopen Codira.

Monthly credits arrive once per calendar month (UTC), usually the first time the app checks your account that month, so open the app. Contact support@codira.com if they're missing.

You've hit your plan's per-minute limit: Hobby 60, Core and Core+ 200, Ultra 300, Teams 500, Enterprise 1,000 requests a minute (see Plans and pricing). Wait a moment and try again.

The model you picked isn't included in your plan, or isn't one Codira's gateway offers. Choose another model in Settings → AI Models, upgrade, or use your own API key for that provider.

Brief network problems are retried automatically. A reply that goes silent for 2 minutes is stopped. Try again. If the Composer seems stuck, press Esc or Stop, check your credits or keys, and try a simpler request. If the same kind of request keeps hanging, email support@codira.com with the prompt.

Your saved OpenAI key is wrong or has expired. Go to Settings → API Keys and click Clear for OpenAI. Codira then uses your credits instead.

Usually because:

• it reached its daily limit (turn it back on to resume)
• the charge would go over your monthly spending cap
• there's no default card
• the card was declined

The last error is shown on Dashboard → Credits.

Raise the cap on Dashboard → Credits (from $50 up to $5,000; the default is $200), then try again. The total can take up to a minute to update.

If you use your own API keys, nothing changes. On Codira credits, AI features pause until service returns. Your code isn't affected either way. Check codira.com/status.

Open the Canvas tab. Preview Doctor checks before it starts your app and diagnoses it if it fails:

• Missing dependencies: one-click install with npm, yarn or pnpm.
• Port already in use: Free port, which names the program using it and stops it.
• OpenSSL "digital envelope" error: retries with the legacy setting.
• A dev server already running in the terminal: the preview connects to it instead of starting a second one.

Codira also looks for apps in subfolders such as frontend/, web/, apps/* and packages/*. Use ↗ Browser to open the preview in your normal browser with full developer tools.

It needs your dev server running. Start the preview and it will run after your next applied change.

That's on purpose. Codira's editor shows syntax errors only, because without your full project setup it would mark correct code as wrong. Run your project's type check in the terminal, for example npm run typecheck. Codira's agents also check the build after every change they apply.

No. Deleting in the Explorer is permanent. Time Machine can only bring a file back if it was captured in a checkpoint before an AI change.

This was fixed long ago. Update Codira. On a very old version, click Stop, check that the folder exists, then use Open Folder.

Click Show details for the build log, or Fix it with AI, which uses your credits, then press Update. Common causes are a missing build or start command, or a server that doesn't listen on the PORT it's given. A failed update never takes down the version that's already live.

Sync pulls the latest commits and replays yours on top. When a file conflicts, the Ship panel switches to a "stopped sync" view, and conflicted files show a red !. Then:

1. Open each conflicted file and remove the <<<<<<<, ======= and >>>>>>> markers, keeping the code you want.
2. Click Finish sync.

To back out instead, click Undo sync, which puts everything back exactly as it was. Codira won't commit a file that still contains conflict markers. There's no visual merge tool; you fix conflicts in the editor.

The message names the step that failed (stage, commit or push) and suggests a fix. If it's a sign-in problem, check git or gh in the terminal.

Codira uses your own git credentials and never asks for a password. Set up SSH keys or run gh auth login in the terminal, then try again.

Codira loads your shell setup from .zprofile, not .zshrc. If your Homebrew or nvm lines are only in .zshrc, move the brew shellenv line to .zprofile, then restart Codira.

Codira uses whichever account gh is signed in with. Run gh auth login or gh auth switch in the terminal.

Click Initialize repository in Source Control or Time Machine. The first time you apply an AI change, Codira also sets up git itself and makes a baseline commit ("Initial commit (Codira baseline)"), because Time Machine needs it.

Codira checks the remote in the background every 60 seconds and when you return to the window. When you're offline, those checks fail quietly, so reconnect and open the Ship panel.

Codira needs git, and checks for it at launch. It needs Node.js and npm for previews and JavaScript builds, and the GitHub CLI (gh) for creating repos and pull requests. It shows you how to install whatever's missing:

• Mac developer tools: Run this for me starts Apple's installer.
• git or `gh` with Homebrew: a button runs brew install in the terminal.
• Node.js, or no Homebrew: Codira opens the official download page.

After installing, quit and reopen Codira so it can find the new tools.

When macOS asks, choose Always Allow.

There's no reset button. To start fresh on a Mac:

1. Quit Codira (⌘Q).
2. Delete the folder ~/Library/Application Support/dev.codira.ide. This removes your settings and chat history.
3. Optionally, open Keychain Access and delete the items that start with dev.codira.ide (your API keys, sign-in, Jira token and environment variables).
4. Optionally, delete the .codira folder in a project to clear that project's memory, index and plans. Codira rebuilds what it needs.

• Mac: quit Codira and drag it from Applications to the Trash. Then follow steps 2 and 3 above to remove your data.
• Windows: use Settings → Apps → Installed apps (Apps & features).

Codira doesn't write a log file, and there's no "send logs" button yet. Instead, email support@codira.com with:

• your Codira version (Settings → About Codira)
• your operating system and its version
• the steps that cause the problem
• the exact error text, or a screenshot

Crash reports reach Codira automatically unless you've turned them off. For Codira hosting, build and live logs are on the app's page under Dashboard → Hosting.

• Email: support@codira.com. Include your Codira version (Settings → About Codira), your operating system, the steps to reproduce the problem and the error text.
• Billing: email support@codira.com with "billing" in the subject line, which sends it to a separate queue.
• Bugs: Dashboard → Bug reports on codira.com.
• Security issues: security@codira.com.

Product support and billing

support@codira.com

General questions

hello@codira.com

Security and vulnerability reports

security@codira.com

Privacy and data requests

privacy@codira.com

Legal

legal@codira.com

Sales and Enterprise

sales@codira.com

Partnerships

partners@codira.com

Press

press@codira.com, media@codira.com

Careers

careers@codira.com

Copyright (DMCA)

copyright@codira.com

Abuse on Codira hosting

abuse@codira.space

Go to Dashboard → Bug reports on codira.com. Give:

• a title (at least 4 characters)
• a description (at least 10 characters)
• a severity: low, medium, high or critical

You can add steps to reproduce, your environment, a URL and up to 8 screenshot links. Only you and Codira's team can see your reports; they aren't shared with your workspace. The team replies in the report's comment thread. Status moves from Open to Triaged to In progress, then Fixed, Won't fix or Duplicate. Send security issues to security@codira.com instead.

Yes, at codira.com/status. It covers the website, sign-in and accounts, the AI service and app updates, and Codira hosting (hosted apps and launching). Each has 90-day uptime bars and the last 14 days of incidents, in UTC.

• Checks run every 5 minutes.
• Two failed checks in a row count as an outage. A check that takes longer than 5 seconds counts as degraded.
• The AI service also follows Anthropic's and OpenAI's own status pages.

1.6.9 on Mac. codira.com/download always gives you the newest available installer for your computer.

• Terms of Service: codira.com/terms
• Privacy Policy: codira.com/privacy
• Data Policy: codira.com/data-policy
• Acceptable Use Policy: codira.com/legal/acceptable-use
• Copyright (DMCA): codira.com/legal/dmca
• Security: codira.com/security
• Trust: codira.com/trust

You do. You keep ownership of what you put in, and Codira gives you any rights it has in what the AI produces. There's no Codira branding in your code.

No. AI output is provided as is, and you're responsible for reviewing and testing it. Codira's liability is capped at the greater of 6 months' fees or $100.

Reverse engineer it, resell it, use it to build competing AI models or datasets, scrape it, probe its security, or submit regulated data without authorisation. The Acceptable Use Policy also covers what you can host on codira.space, including phishing, malware, scams and impersonation.

Send a DMCA notice to the Copyright Agent, Codira, Inc., 6166 North Scottsdale Road, C4003, Paradise Valley, AZ 85253, or email copyright@codira.com. Counter-notices are accepted at the same address.

Codira, Inc.

• Bryan Ferre, Founder and CEO
• James K. Farrelly, President and COO
• Christian Brotherson, Chief Product Officer
• Robert Clawson, Business Development and Investor Relations

No. The core product is closed source. Some pieces, such as prompt templates, may be open-sourced later.

Yes. codira.com/investors is a private portal for approved investors. You can request access through its sign-up page, and the team approves each account. Investor questions go to Robert Clawson's team through that portal.