# Behest Developer Docs — LLM Index

> Behest is an inference gateway with per-end-user JWT auth, session memory, threads, and BYOK. OpenAI-compatible. Safe to call from a React component.

## Base URLs
- Mint endpoint (control plane): POST https://api.behest.app/v1/auth/mint
- JWKS (control plane): https://api.behest.app/.well-known/jwks.json
- Chat completions (data plane): POST https://{slug}.behest.app/v1/chat/completions
- Threads (data plane): https://{slug}.behest.app/v1/threads
- JWT `iss` claim value: "https://api.behest.app"

## Core reference
- [Overview](/docs/index.md) — Map of all docs
- [Core Concepts](/docs/concepts.md) — Tenants, projects, slugs, sessions, threads, tiers
- [Authentication](/docs/authentication.md) — API keys, mint, JWT, JWKS, local signing
- [API Reference](/docs/api-reference.md) — All endpoints

## Quickstarts (5-min)
- [React + Vite](/docs/quickstarts/react-vite.md)
- [Next.js App Router](/docs/quickstarts/nextjs-app-router.md)
- [Vercel Edge + AI SDK](/docs/quickstarts/vercel-edge.md)
- [Supabase Edge Functions](/docs/quickstarts/supabase-edge.md)
- [Lovable + Supabase](/docs/quickstarts/lovable-supabase.md)
- [Modelence](/docs/quickstarts/modelence.md)
- [Python FastAPI](/docs/quickstarts/python-fastapi.md)
- [Node + Express](/docs/quickstarts/node-express.md)

## Guides
- [Auth modes](/docs/guides/auth-modes.md) — API-key mint vs local signing; when to use each
- [Tiers and usage](/docs/guides/tiers-and-usage.md) — Free/Pro/Max, /v1/billing/usage, 402 handling
- [Multi-conversation chat](/docs/guides/multi-conversation-chat.md) — Sessions + threads end-to-end
- [Streaming UI](/docs/guides/streaming-ui.md) — SSE, AbortSignal, reconnect
- [Error handling](/docs/guides/error-handling.md) — 401/402/429/5xx, typed SDK errors
- [Migrating from OpenAI](/docs/guides/migrating-from-openai.md)
- [Migrating from OpenRouter](/docs/guides/migrating-from-openrouter.md)

## Integrations (auth provider → Behest uid)
- [Supabase](/docs/integrations/supabase.md)
- [Clerk](/docs/integrations/clerk.md)
- [Auth0](/docs/integrations/auth0.md)
- [NextAuth](/docs/integrations/nextauth.md)
- [Firebase Auth](/docs/integrations/firebase.md)

## SDKs
- [TypeScript](/docs/sdk-typescript.md) — `@behest/client-ts`
- [Python](/docs/sdk-python.md) — OpenAI Python + Behest headers

## Platform features
- [Providers / BYOK](/docs/providers.md)
- [Models](/docs/models.md)
- [Guardrails](/docs/guardrails.md)
- [Rate limiting](/docs/rate-limiting.md)
- [Webhooks / events](/docs/webhooks-events.md)

## AI-tool prompts (paste into Lovable/Cursor/Claude/v0)
- [Lovable](/docs/prompts/lovable.md)
- [Cursor](/docs/prompts/cursor.md)
- [Claude](/docs/prompts/claude.md)
- [v0](/docs/prompts/v0.md)

## Key endpoints
- `POST /v1/auth/mint` — Exchange API key for JWT. Body: `{user_id, ttl?, tier?, session_id?}`.
- `POST /v1/chat/completions` — OpenAI-compatible. Headers: `Authorization: Bearer <JWT>`, optional `X-Session-Id`, `X-Thread-Id`.
- `GET /v1/threads` — List threads for current uid.
- `GET /v1/threads/{id}/messages` — Fetch messages.
- `DELETE /v1/threads/{id}` — Delete thread.
- `GET /v1/billing/usage` — Platform + BYOK buckets.

## Auth modes
1. **API-key mint**: server holds `behest_sk_live_...`, calls `POST /v1/auth/mint` per user, returns JWT to client. Default.
2. **Local signing**: server holds tenant-scoped RS256 private key, mints JWT locally with same claims. No network hop per user; use when your backend is hot path.

## Tiers
- `free` — small quota, good for demos
- `pro` — production defaults, higher RPM + token budget
- `max` — highest limits; tune `sort_order` overrides in dashboard.

## Key rules for LLM code generation
- Never expose `behest_sk_live_*` to the browser; mint on server.
- Always pass `user_id` when minting so rate limits and memory are per-user.
- Use `X-Session-Id` or JWT `sid` claim for per-session memory. No header → "default" session.
- Use `X-Thread-Id` for persistent conversations with full message history.
- 402 = over tier quota; show upgrade UI; do not retry.
- 429 = rate limit; respect `Retry-After`.