Skip to content

ModelBridge · Multi-Protocol API Reference

5 endpoint groups

Interactive specification covering OpenAI, Anthropic, and Gemini client ingress, /api/* admin controls, Node.js 24 SQLite /_modelbridge/state RPC, and platform budgets.

POSTAuthorization: Bearer <client-api-key>Core Client Ingress

OpenAI Ingress Endpoints

POST /v1/chat/completions · /v1/completions · /v1/embeddings · GET /v1/models

Resolves routes across upstream:<id>:<model>, @cf/ prefixes, custom modelMap, and 65 preset aliases, failing over across at most 3 accounts strictly on 401/403/404/429.

View routing precedence & credential domains ↗

✅ Capabilities

  • Preserves tool calling, strict schemas, and signed thinking replay metadata across providers
  • Enforces per-platform budgets: Cloudflare 16 MiB, Makers 1 MiB, ESA 16 MiB / 4 fetches / 9s TTFB / 115s total

⛔ Explicit Limits / Rejections

  • Does not implement /v1/responses, image generation/editing, or audio TTS/STT
  • modelMap is not an access-control allowlist; timeouts and 5xx errors never trigger duplicate retries

Parameters, Credentials & Budgets

Authorization / modelheaderBearer / stringRequired

Client key and model slug (supports upstream:<accountId>:<model>).

Status & Semantic Pre-Flight

200 OKStandard OpenAI response or SSE chunks