Understanding the free tier

What you get

• 250 LLM calls per 30-day rolling window across all providers in this workspace.
• 1 seat: the workspace can have exactly one member. Invites beyond seat 1 are blocked at invite time and re-verified at accept time.
• BYOK for ongoing use: new accounts get 7 platform-funded calls (Critique, Improve, Rewrite on Sonnet 4.6) to explore the workbench, then bring your own provider key. See the callout above.
• No public API: the REST API is Solo tier and above. You can generate an API key on Free, but every call returns a 403 Forbidden error.
• Editor, AI pair, evaluation suites, regression testing, 7-day version view, import/export: all available on Free. Full version history requires Solo tier or higher.

Checking remaining quota

Open the Usage or Billing page for the current period's call count and remaining quota. The number is computed on every page load, so it's always within a few seconds of real-time.

7-day version history view

On the Free tier, the version history panel, search, annotations, and dashboard activity feed only show prompt versions created in the last 7 days. This is a view filter, not a retention policy: every version you've ever created is still stored in the database. When you upgrade to Solo or higher, the 7-day filter drops off on your next page load and the full history reappears instantly. You don't lose anything by staying on Free for a while; you just can't browse or diff older versions until you upgrade.

The filter applies to reads only. You can still create new versions at any rate (every edit, AI suggestion, or restore creates a version), and the latest current_content on every prompt is always visible regardless of age. Restoring or branching from a version older than 7 days is blocked on Free with a clear "hidden on Free plan" error.

Evaluation suites

Free workspaces can author, run, and judge evaluation suites end-to-end with LLM-as-a-judge scoring. Each judge call counts as one of your 250 monthly AI calls; a 50-case suite that judges every case burns 50 calls. The 250-call cap is the natural cost lever, not a feature gate. See Evaluation suites.