Skip to content

CONFIGURATION.md

Latest commit

 

History

History
75 lines (53 loc) · 2.42 KB

CONFIGURATION.md

File metadata and controls

75 lines (53 loc) · 2.42 KB
Raw

Configuration

LLM Router reads configuration from environment variables. A local .env file in the project root is loaded automatically.

Required

Variable Description
UPSTREAM_BASE_URL OpenAI-compatible upstream base URL, without a trailing slash.
UPSTREAM_API_KEY API key for the upstream relay.

Optional

Variable Default Description
PORT 8787 Local HTTP port.
ROUTER_API_KEY empty If set, clients must call the router with Authorization: Bearer <ROUTER_API_KEY>.
UPSTREAM_TIMEOUT_MS 30000 Timeout for upstream requests.
AUTO_MAX_ATTEMPTS 2 Maximum answer-model attempts for auto requests.
LLM_ROUTER_CONFIG_PATH .llm-router.local.json Local runtime config file used by the Admin page.

Example

PORT=8787
UPSTREAM_BASE_URL=https://your-relay.example.com
UPSTREAM_API_KEY=sk-your-upstream-key
ROUTER_API_KEY=local-router-key
UPSTREAM_TIMEOUT_MS=30000
AUTO_MAX_ATTEMPTS=2
LLM_ROUTER_CONFIG_PATH=.llm-router.local.json

Local Admin page

Open the local Admin page after the server starts:

http://127.0.0.1:8787/admin

The page lets you inspect upstream models and choose which model is used for the first routing call of auto, auto-coding, and auto-longtext.

  • automatic: use the cheapest model with known pricing as the router model.
  • manual: use the selected upstream model as the router model, even if its price is unknown.

The router model still only returns routing JSON. The final answer is always a second independent call to the selected target model.

If ROUTER_API_KEY is set, the Admin page shell can still be opened locally, but config reads and writes require the same key. The page stores that key only in the browser's localStorage.

The config API is also available directly:

curl http://127.0.0.1:8787/admin/config \
  -H "Authorization: Bearer local-router-key"

curl -X POST http://127.0.0.1:8787/admin/config \
  -H "content-type: application/json" \
  -H "Authorization: Bearer local-router-key" \
  -d '{"router_model_id":"gpt-5.5"}'

Use {"router_model_id": null} to return to automatic router-model selection.

Client configuration

Use the router as the OpenAI-compatible endpoint:

base_url=http://127.0.0.1:8787/v1
api_key=local-router-key
model=auto

If ROUTER_API_KEY is empty, the client API key can be any placeholder value.