AI limits

What counts as a request

Each run you start in the agent panel is one AI request, in any mode. Tokens sent to and received from the provider are counted separately. Free: 50 requests and 500k tokens per month; Pro: 1,000 and 10M; Team: 5,000 and 50M. Counters reset monthly.

The run was refused with quota_exceeded

The monthly request or token allowance is used. Wait for the reset, or upgrade from Workspace → Billing. The error says which limit (ai_request or ai_tokens).

The run stopped early

Runs are bounded: 40 tool calls, 60 model turns, 25 minutes, 2 repair attempts. The final message says which bound was hit. Split the task into smaller requests; see prompting.

"AI is not configured"

The deployment has no Anthropic or OpenAI key on the server. The operator must configure one; nothing in your account changes this. Templates, editor, previews and deployments work without AI.

Which model is used?

The agent panel shows the active provider and model. The operator chooses the default; per-workspace selection is not available yet.

Is my code sent to the provider?

Yes, the relevant files and your messages are sent to the configured provider for each run. Environment variable values are never sent. See the AI policy.