Error codes

Fire Mission preserves the upstream provider's error envelope shape — OpenAI on /v1/chat/completions, /v1/embeddings, /v1/models; Anthropic on /v1/messages. Inside that envelope, error.code may be one of the Fire Mission-specific codes below.

Envelope shapes

OpenAI surface:

{ "error": { "message": "...", "type": "...", "code": "...", "param": null } }

Anthropic surface:

{ "type": "error", "error": { "type": "...", "message": "..." } }

Fire Mission codes

Code	HTTP	Surface	Meaning	Remediation
`monthly_cap_exceeded`	429	both	Free-tier monthly call cap reached.	Upgrade at /pricing for unlimited calls.
`spend_cap_exceeded`	429	both	Per-key daily spend cap reached.	Raise the cap in the dashboard or wait for the UTC day rollover.
`spend_cap_unavailable`	503	both	Spend cap is enforced but the tally service is temporarily unavailable.	Retry shortly. Fail-closed by design — no uncapped traffic.
`rate_limit_exceeded`	429	both	Per-key burst rate limit exceeded.	Back off and retry with exponential jitter.
`provider_blocked`	403	both	Provider blocked by US-compliance / supply-chain policy.	Configure a different BYOK provider on the supply-chain whitelist.
`security_violation`	400	both	PII, prompt-injection, or API-key-leakage detected in the request.	Sanitize the input. Categorical detection codes only — no matched substrings are ever returned or logged.
`model_substitution_detected`	502	both	The responding model id did not match the requested model.	Switch to a verified provider or contact support.
`model_not_found`	400	both	No configured BYOK provider can serve the requested model.	Configure a provider that supplies the model in the dashboard.
`wrong_endpoint`	400	both	The model is served on a different endpoint (e.g. Claude on /v1/messages).	Use the SDK that matches the model family, or call the right endpoint directly.
`moe_routing_failed`	503	both	MoE routing could not produce a viable decision.	Drop the `x-fm-route` header to bypass MoE, or pick a specific model.
`moe_routing_unavailable`	503	anthropic	MoE produced a non-Anthropic-shape model on the /v1/messages surface.	Constrain MoE to Anthropic-shape providers or use /v1/chat/completions.

Standard upstream codes

Anything not in the table above is a pass-through from the upstream provider (e.g. OpenAI's insufficient_quota, Anthropic's overloaded_error). These reach you unchanged.