Hobby
$0
free forever
- No credit card required
- Limited agent requests
- Limited inline (tab) completions
- Base models
Individual
$20
per month
Everything in Hobby, plus:
- 4,000 credits / month
- Frontier models
- The 10 verified agents — plan → implement → review → security → QA
- Genesis, Sidekick & the live Canvas preview
- MCPs, skills, and hooks
- Framework-agnostic UAT (every change ships verified)
- Unlimited Architect + 30-day Time Machine
- Automatic credit top-up
Teams
$40
per seat / month
Everything in Individual, plus:
- 8,000 credits / month / seat
- Centralized team billing and administration
- Shared team memory + agent presets
- Usage analytics to understand team behavior
- Team-wide privacy mode
- SAML / OIDC SSO
- Audit logs
- Priority support
1 × $40 = $40/mo
Enterprise
Custom
annual contract
Everything in Teams, plus:
- Pooled usage to maximize budget efficiency
- Invoice / PO billing
- SCIM seat management
- Repository, model, and MCP access controls
- Auto-run, browser, and network controls
- Audit logs and service accounts
- AI code tracking API
- Priority support and account management
Enterprise contracts. Custom infrastructure.
For organizations 50+ seats, or any team with specific compliance, security, or deployment requirements. Annual contract, custom MSA available.
What’s included
- Unlimited credits or custom allocation
- Unlimited parallel agents per seat
- Advanced SSO — SAML 2.0, SCIM, JIT
- SOC 2 Type II report (under NDA)
- DPA with EU SCCs + BAA for HIPAA
- Self-hosted gateway option
- Dedicated CSM + Slack Connect
- 4-hour business-hours SLA, 24/7 critical
- Custom MSA, liability caps, data residency
- Onboarding workshop included
How it works
Tell us the team shape, the compliance bar, and the deployment footprint. We’ll put together a contract that fits. Pricing is typically a flat annual commit with usage thresholds — no per-seat micro-counting.
Comparison
| Feature | Hobby | Individual | Teams | Enterprise |
|---|---|---|---|---|
| Price | Free | from $20/mo | from $40/seat/mo | Custom |
| Bundled credits / month | 200 | 4,000 – 50,000 | 8,000 – 20,000 / seat | Custom |
| Buy additional credits | ✓ (one invoice) | |||
| Auto-topup | ||||
| BYO API keys | ||||
| Workhorse models | ||||
| Frontier models | ||||
| The 10 verified agents | limited | |||
| Genesis, Sidekick & live Canvas | ||||
| MCPs, skills & hooks | ||||
| Framework-agnostic UAT | ||||
| Full Review mode | 5/mo | unlimited | unlimited | unlimited |
| Architect (visual stack/schema) | 1 project | unlimited | unlimited + shared | unlimited + shared |
| Time Machine | 7 days | 30 days | unlimited | unlimited |
| Patch Guards + auto-updates | ||||
| Parallel agents | 1 | 4 – 8 | 6 – 8 / seat | unlimited |
| Shared team memory & presets | ||||
| Usage analytics | ||||
| Team admin & centralized billing | ||||
| Audit logs | Unlimited + export | |||
| SAML / OIDC SSO | ||||
| Advanced SSO (SCIM, custom IdP) | ||||
| Repository / model / MCP access controls | ||||
| Auto-run, browser & network controls | ||||
| SOC 2 Type II report | ||||
| DPA / BAA | ||||
| AI code tracking API | ||||
| Support | Community | Priority | Dedicated CSM · 4h SLA | |
| Card required | No | Yes | Yes | Custom |
One agent run. 5–15 credits.
One agent run = one full plan → implement cycle. That’s the planner + implementer agents firing once each. Cost varies with the model the gateway routes you to and how much context the agents need.
01
Full Review mode is heavier.
It adds the reviewer + security + QA agents on top, so a full-review run costs roughly 3× a basic run.
02
Credits roll over for 90 days.
Bundled monthly credits expire first, purchased credits last — so paying for more credits never penalizes you for taking a break.
03
Auto-topup never refuses a request.
Set a daily cap and a monthly ceiling in your dashboard. The gateway tops up automatically when you’re below your threshold mid-flow.
04
BYO keys cost zero credits.
Paste your own OpenAI or Anthropic key in Settings. You bypass the gateway entirely and pay the provider directly — every Codira feature still works.
Buy more credits anytime.
One-time packs. Purchased credits never expire as long as your account is active. Larger packs get a better per-credit rate.
Start with everything. Decide after.
7 days of full Pro
Every new account starts with a 7-day Pro trial — every model (frontier included), unlimited Full Review, unlimited Architect and Time Machine, and 1,000 gateway credits to start. No card to begin. It’s the real product, not a stripped-down demo.
When the trial ends
Pick a plan to keep the ai features going — upgrade in-IDE from Settings → Account. Your projects, Architect docs, and Time Machine history stay exactly where they are; nothing is deleted. Prefer to run on your own keys? BYO is included on every paid plan.
To prevent abuse, new accounts require email verification before the gateway will route any requests, and the trial is one per user. We rate-limit per device fingerprint and per IP — accounts that look automated get blocked at the gateway with no impact on real users.
Frequently asked questions.
For anything not covered here, write to hello@codira.com.
The basics
Cursor and Copilot are autocomplete with a chat sidebar — one model writes code and you decide if it’s right. Codira routes every change through ten specialized agents (planner, implementer, reviewer, security, QA, auditor, verifier, explainer, debugger, UX reviewer) plus six deterministic safety checks before you ever see a diff. Same output format as Cursor, fundamentally different process. Side-by-side comparison →
Working developers who ship production code and want ai that actually helps instead of creating new debugging work. Also good for solo founders moving fast and teams that want consistent quality from ai-generated changes. It’s less of a fit for casual “let me try ai” exploration — for that, Cursor’s free tier is easier.
Yes. Codira makes existing developers faster and safer; it doesn’t generate working software for non-developers. If you can’t read the diffs Codira proposes, you can’t catch the rare bugs it introduces.
Ten specialized agents, each with a defined job: plan it, implement it, review it, scan it for security issues, generate tests, audit existing code, cross-check those findings, explain unfamiliar code, debug failures, and design-review the rendered UI against your design system. You watch them work live in the chat panel. If any one of them rejects the change, the run goes back to the implementer with structured feedback.
Privacy & data
Depends on routing. With BYO API keys, your code goes directly to OpenAI/Anthropic — Codira never sees it. With the hosted gateway, your code passes through Codira’s servers on its way to the same providers; we forward the request, log token counts for billing, and don’t retain the request body. Either way, we never train on your code.
No. Codira doesn’t have its own models — we proxy to OpenAI and Anthropic, both of which contractually don’t train on API traffic. We also don’t store your code on our servers regardless of which provider routes it.
Locally: your project files, conversation history, Architect docs, Time Machine checkpoints, settings, BYO API keys (stored in
~/Library/Application Support/dev.codira.ide/settings.json, chmod 0600; macOS Keychain migration is on the roadmap).On our servers (only if signed in): account info, subscription status, credit wallet balance, transaction ledger.Your code never lives on our servers.SOC 2 Type II is planned for the v1.0 launch window — we’re working on the controls and observation period now. Report will be available under NDA on the Enterprise plan once complete. If your compliance team needs documentation before then, contact security@codira.com and we’ll work through your requirements directly.
The IDE already runs locally on your machine — nothing to self-host for the editor itself. The hosted gateway (the proxy for credit-based billing) is the only piece running on our infrastructure. Enterprise customers can self-host the gateway and point Codira at their own deployment, keeping all ai traffic on their network.
Pricing & billing
Every new account gets 7 days of full Pro access — every model, unlimited Full Review, Architect and Time Machine, plus 1,000 gateway credits to start. No card to begin, and no automatic charge: when the trial ends, you pick a plan to keep the ai features going. Your projects and history stay put either way. One trial per user.
A unit of usage on Codira’s hosted gateway. One credit ≈ $0.01 retail. A typical agent run costs 5–15 credits depending on which model the gateway routes you to and how much context the agents need. With BYO keys you skip credits entirely — you pay the API provider directly and Codira doesn’t see the bill.
Two safeguards keep you from losing partial work. First, Codira reserves an estimated amount up front before a multi-step run starts — if you don’t have enough, you’re told before any ai call fires. Second, you can enable auto-topup so the gateway re-bills your card before you hit zero.
Anytime, no restart needed. Settings → API Keys shows live routing per provider; pasting a BYO key automatically overrides the gateway for that provider, clearing it falls back to the gateway. Many users keep Anthropic BYO (cheaper for them) and OpenAI on the gateway.
Annual billing saves ~8% (pay for 11 months, get 12) on Individual and Teams — toggle Monthly/Annual at the top of this page. Volume discounts on Teams kick in at 10 / 25 / 50 seats (10% / 15% / 20% off respectively). Enterprise is custom.
Within 14 days of your first paid charge: full refund, no questions. After that: prorated refunds for downgrades; subscription cancellations take effect at the end of your billing period (no surprise charges, no service cutoff mid-month).
Compatibility & install
macOS and Windows today. The macOS app is a universal binary — one DMG runs natively on both Apple Silicon (M1/M2/M3/M4) and Intel Macs, macOS 11 (Big Sur) or later. The Windows build is a signed installer for Windows 10 and 11 (64-bit). Linux is on the roadmap but not committed to a date.
Yes. Codira doesn’t conflict with other editors — they can all open the same project. Many users start by running Codira for the ai-heavy work and keeping their existing editor for habits they don’t want to change. Most migrate fully over time.
No. Time Machine checkpoints live in a parallel namespace (
refs/codira/cp/*) that doesn’t appear in git log, doesn’t push to GitHub by default, and doesn’t pollute your branches. The only commits Codira makes on your real branches are the ones you explicitly accept by applying a patch.Codira works on any text-based codebase, but the agents are most effective on mainstream stacks where the underlying models have strong training data — JavaScript / TypeScript, Python, Go, Rust, Ruby, Swift, Java. Framework helpers exist for Next, Vite, Svelte, Remix, Astro, FastAPI. Niche languages work but you’ll want to set up Architect with your stack so the planner has context.
Features & limits
You’ll see it in the diff and can reject the patch — no changes touch your files until you click Apply. If a bad change gets through Review and you only catch it after applying, Time Machine has a one-click rewind on every ai run. The deterministic patch guards catch the most common failure modes (fabricated code, sneaky deletions, unrequested rewrites) before the patch reaches you.
Yes. The Agents panel (sparkles icon in the activity bar) has toggles for the optional agents — Correctness Reviewer, Security, QA, and UX Reviewer — that you can flip on or off per workspace. The Planner and Implementer always run. The chat panel slash commands today are
/plan, /remember, and /analyze. Per-agent slash commands (/review, /security, /qa on existing patches) are roadmap.Editor, terminal, file browser, git: yes. Anything ai: no — we don’t ship local models. If your internet drops mid-run, the in-flight request fails cleanly and you can retry when you’re back online.
Codira indexes the workspace incrementally and only reads files when an agent actually needs them, so it scales to large repos. Internal testing has gone up to ~200k-LOC codebases without issues. Monorepos work — point Codira at the root and it’ll respect your
.gitignore for skip-dirs.Trust & support
Submit it from the in-app dashboard at Dashboard → Bug reports. You’ll get a thread we can reply to directly, status updates as we triage, and a permanent record you can come back to. For security issues or anything sensitive, email security@codira.com instead. Response time depends on your tier — community (best-effort), Solo 72h, Pro 48h, Team 24h, Enterprise 4h SLA.
The IDE itself isn’t, today. Pieces of it may open-source over time (the agent prompt library, the patch parser, the deterministic guards) as the product stabilizes — open-sourcing closer to launch tends to fragment the user base before it has a chance to settle.
Near-term: a Linux port, a real model-picker UI, team-shared agent presets, and Architect → direct code generation. Longer-term: real-time pair-programming mode, a plugin API for custom agents. Public roadmap at codira.com/roadmap once we commit dates.
Probably, eventually — the ai provider costs that feed credit pricing keep changing. But anyone subscribed at the time of a price change is grandfathered for the life of their subscription. We won’t pull a “your current plan is being deprecated, please upgrade to keep your features.”
Try Codira. Decide later.
The full IDE downloads in seconds. Pricing scales when you’re ready.