Errors

If the turn never starts, you get an HTTP error. If it starts and then fails, you get an error event on the stream. HTTP body:

{ "error": { "code": "rate_limited", "message": "…" } }

On a stream error event, retryable: true means a model or provider blip — retrying is reasonable. retryable: false means the request was wrong — fix it.

Code	HTTP	Meaning
`invalid_request`	400, 422	Bad body, or `context` sent on a continue turn.
`unauthorized`	401	Bad, expired, or revoked API key or client token.
`insufficient_credits`	402	No credits left.
`conversation_not_found`	404	Wrong id, or a token asking for a conversation it is not bound to.
`token_not_found`	404	Revoke was called for a token this vendor does not own.
`turn_conflict`	409	A turn is already running on this conversation. Wait for it to finish.
`rate_limited`	429	Per-token or per-vendor cap. Check `Retry-After`.
`model_unavailable`	503	All model providers unavailable. Retryable.
`service_unavailable`	503	Token store is temporarily unavailable. Retry.
`internal`	500	Server error.

See rate limits for 429 caps, and client tokens for 401 / 404 on tokens.

⌘I

Getting Started

Make a Search

Get Product Data

Conversational Agent

Checkout

Pricing, Reporting, and Commissions

Build

Integrate

FAQ