Wanaku

Appearance

You are reading the docs for Wanaku next, the Rust-based server (early access of 0.3.0). Read the Classic docs for the Java-based Wanaku 0.2.

Proxy OpenAI-compatible inference traffic ​

The inference proxy is a raw, transparent reverse proxy (port 8083) to an OpenAI-compatible backend. It forwards requests as-is, including the caller's own Authorization header — Wanaku does not inject or store a credential for it. Point it at a backend with WANAKU_INFERENCE_UPSTREAM:

bash
export WANAKU_INFERENCE_UPSTREAM=127.0.0.1:11434

The Wanaku Debugger plugin provides the LLM Chat page. Install the plugin from the Admin UI Plugin Catalog. Open Debugger from the Developer navigation section. Select the LLM Chat tab. The page calls this port directly with a key you supply in the browser. See Management API for the endpoint shape.

By default, the LLM Chat page keeps the API key in memory for the current session only. The page does not save the key to local storage. To keep the key across sessions, first enable Remember LLM settings in this browser. Then enable Also remember the inference API key. The page writes the key to browser local storage. A script or a browser extension on the page can then read the key. Enable this option only on a trusted device.

Because the proxy forwards the Origin header unchanged, the backend's own origin policy still applies. Ollama, for example, rejects browser-origin requests by default — set OLLAMA_ORIGINS on the Ollama side to allow the Admin UI's origin. Wanaku's own CORS filter on port 8083 controls only the response back to the browser; it does not affect what the backend accepts.

Wanaku also rewrites the outgoing Host header to match WANAKU_INFERENCE_UPSTREAM, instead of forwarding the browser's original Host. Backends that route by hostname (for example, a TLS-terminating ingress in front of a hosted LLM API) reject a mismatched Host. Note: like WANAKU_CORS_ORIGIN, this override applies only to the embedded default pipeline config. A custom --pipeline-config must define its own headers filter with request_set on the inference_proxy chain to get this behavior.

WANAKU_INFERENCE_UPSTREAM may include a path, for example https://openrouter.ai/api. Wanaku prepends that path to every request it forwards — a request to /v1/chat/completions on port 8083 reaches https://openrouter.ai/api/v1/chat/completions upstream. The same custom---pipeline-config caveat applies: this requires a path_rewrite filter with add_prefix on the inference_proxy chain.

Next steps ​