llmleaf is a llm proxy. It proxies different llm providers and their slighty different apis and converts it to a single api surface (enhanced openai-compatible or anthropic).

• fast
• efficient
• light-weight
• extensible

• One stable endpoint in front of every provider — consumers speak OpenAI, OpenRouter, or Anthropic dialects; llmleaf maps them to one internal model and back.
• Streaming-first (SSE): settings.upstream_streaming = "always" is the default, so chat calls use the provider's incremental upstream transport even when the consumer requests a non-streaming response. Set the per-provider policy to "when_requested" or "never" to opt out.
• Modalities: chat, embeddings, rerank, text-to-speech, speech-to-text, realtime (WebSocket), batch jobs.
• Per-model fallback chains with node-local, health-aware switchover — no consensus or shared state, so N nodes run behind a plain load balancer.
• Opt-in per request/provider: Anthropic prompt caching, a unified thinking/reasoning-effort ladder.
• Responses API both ways: consumers can POST /v1/responses to any routed provider — upstreams without a Responses endpoint are served over their chat-completions wire transparently. Upstream, OpenAI and xAI speak their Responses APIs by default; OpenRouter's beta POST /responses (signed open-reasoning replay, routed cost), Groq's beta POST /responses (open unsigned reasoning), and Azure OpenAI's v1 surface (resource-scoped POST /openai/v1/responses) are per-provider opt-ins (chat_api = "responses").
• Auth via HTTP-Basic key tokens (optional OAuth2/JWT); identity, limits, topology, and usage ride an outbound control plane (pull verdicts and provider/route config — diff-reconciled on every refresh — push usage). Fully operable from the config file alone.

The use of the websocket api is recommended, this greatly improves the chances of prompt caching actually working.

• Native dialects: Anthropic, Google Gemini, Vertex AI, Cohere, Ollama, LM Studio.
• OpenAI-wire family: OpenAI, OpenRouter, Requesty, Groq, DeepSeek, xAI (Grok), Mistral, Together, Fireworks, Perplexity, Cerebras, Z.AI (GLM), Moonshot (Kimi), MiniMax, Amazon Bedrock, Hugging Face Inference Providers, DeepInfra, Cloudflare Workers AI, OCI Generative AI, Databricks Model Serving, NVIDIA NIM, Azure OpenAI. Moonshot additionally gets a dedicated provider layer that rewrites tool JSON schemas into the upstream's restricted "moonshot flavored JSON schema" (standard Pydantic/zod output otherwise 400s). Subscription plans ride dedicated kinds where the vendor gives them their own endpoint: zai-coding (GLM Coding Plan, /api/coding/paas/v4) and kimi-coding (Kimi for Coding, api.kimi.com/coding/v1); MiniMax's Token Plan shares the standard endpoint, so minimax-token-plan is an alias of minimax (only the key differs).
• echo for local testing.

# Run with the embedded dev config (echo provider, key `local-dev:s3cret`)
cargo run -p llmleaf

# …or point at your own config
cargo run -p llmleaf -- llmleaf.toml

Copy llmleaf.example.toml, fill in provider credentials (use env:VAR indirection — secrets never live in the file), and pass it as the argument. Container image: docker buildx bake image (listens on :8080). Send a request:

curl localhost:8080/v1/chat/completions \
  -H "Authorization: Bearer $(printf 'local-dev:s3cret' | base64)" \
  -d '{"model":"demo","messages":[{"role":"user","content":"hi"}]}'

Base64 the id:password credential with no trailing newline — use printf (or base64 -w0), not echo. A stray newline is encoded into the value, so the decoded password becomes pw\n and fails the hash check → 401 unknown api key, even when the configured pw_hash is correct.

See llmleaf.example.toml for the full configuration surface (providers, routes, keys, control plane).

Consumer endpoints (OpenAI-compatible unless noted):

Read-only admin (optional token): GET /admin/routes, /admin/health, /admin/keys. Official client SDKs for 6 languages live in clients/.

Two strictly separated planes. The core (data plane) is the proxy; the control plane is reached only outbound — the core pulls identity/verdicts/topology and pushes usage, never the reverse. A pulled topology ([control.topology]) lets the controller also serve provider and route configuration, diffed against the previous pull on every refresh so resources are added, updated, and removed incrementally on top of the immutable config file. See SOUL.md for the full design constitution. To build a compatible controller, see the external control-plane implementation guide.

flowchart LR
  Cons["Consumers<br/>OpenAI · OpenRouter · Anthropic"] --> Surf["Compat surfaces"]
  subgraph Core["llmleaf core — data plane"]
    direction LR
    Surf --> Auth["authenticate"] --> In["map in"] --> Route["route + fallback"] --> Stream["stream"] --> Out["map out"] --> Ev["emit events"]
  end
  Route --> Prov["Providers<br/>compiled-in traits · WASM plugins"]
  Prov --> Up["LLM providers"]
  Ctrl[["Control plane (outbound)"]]
  Auth -. "pull identity / verdicts" .-> Ctrl
  Route -. "pull topology (providers + routes)" .-> Ctrl
  Ev -. "push usage" .-> Ctrl

This project is being developed with AI assistance.

Copyright (C) 2026 Fionn Langhans fionnlanghans@codefionn.eu.

llmleaf and its clients are dual-licensed under either the MIT License or the Apache License 2.0, at your option.