Fire Mission docs

Fire Mission speaks the OpenAI and Anthropic wire formats. Point your existing SDK at https://firemission.us/v1 and change only the API key. Everything else — request shape, response shape, streaming events, error envelope — stays the same as the upstream provider.

hello.py python

from openai import OpenAI
client = OpenAI(api_key="fm_live_...", base_url="https://firemission.us/v1")
print(client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Hello"}]
).choices[0].message.content)

Where to start

Python quickstart — OpenAI and Anthropic SDKs.
Node / TypeScript quickstart.
Go quickstart.
curl quickstart.

API reference

Every endpoint on the public /v1/* surface is described in our OpenAPI 3.1 spec — feed it to Stainless, openapi-typescript, or any code-gen tool. Browse it interactively:

POST /v1/chat/completions — OpenAI chat, streaming, tools, JSON mode, vision.
POST /v1/messages — Anthropic messages.
POST /v1/embeddings — text embeddings.
GET /v1/models — list models reachable on your tier.
POST /api/v1/gateway — universal gateway for non-LLM operations.

Guides

BYOK provider keys — how Fire Mission stores and resolves your provider keys.
Streaming — SSE for chat and messages.
MoE routing — auto-route by cost, speed, or quality.
Security scanning — PII, prompt-injection, and key-leakage detection.

What you don't have to think about

Two-Plane isolation. Fire Mission never persists raw prompt or completion bytes — only metadata (token counts, cost cents, status, latency, security verdicts).
BYOK encryption. Your provider keys are AES-256 encrypted at rest and decrypted only on the hot path.
Geopolitical authorization. Every call is checked against the supply-chain whitelist before egress.
Model-substitution detection. If the responding model id doesn't match the requested one, the call fails with model_substitution_detected.