Configuration

LARouter is configured through environment variables, a JSON config file, and runtime API.

Environment Variables

Required (at least one API key)

Variable	Description	Example
OPENAI_API_KEY	OpenAI API key	sk-...
ANTHROPIC_API_KEY	Anthropic API key	sk-ant-...
GOOGLE_API_KEY	Google Gemini API key	AIza...
DEEPSEEK_API_KEY	DeepSeek API key	sk-...

[!IMPORTANT] At least one API key must be configured. For local-only operation with Gemma 4 models, set any key to a dummy value (e.g., GOOGLE_API_KEY=local-only).

Optional

Variable	Default	Description
LAROUTER_PORT	18790	Server port
LAROUTER_HOST	127.0.0.1	Bind address
LAROUTER_ADMIN_KEY	(generated)	Admin API key for dashboard auth
LAROUTER_LOG_LEVEL	info	Logging level: debug, info, warn, error
LAROUTER_DB_PATH	./larouter.db	SQLite database path
LLAMA_CPP_PATH	/usr/local/bin	Path to llama.cpp binaries

Custom Provider URLs

Variable	Default	Description
LAROUTER_OLLAMA_URL	http://localhost:11434	Ollama server
LAROUTER_LLAMACPP_URL	http://localhost:8001	llama.cpp server
LAROUTER_CUSTOM_URL	—	Any OpenAI-compatible endpoint

Config File

backend/config/default.json — default tier mappings and model configuration:

{
  "tiers": {
    "HEARTBEAT": {
      "model": "gemma-4-e2b",
      "provider": "llamacpp",
      "maxTokens": 256
    },
    "SIMPLE": {
      "model": "gemma-4-e4b",
      "provider": "llamacpp",
      "maxTokens": 1024
    },
    "MODERATE": {
      "model": "gemma-4-26b-a4b",
      "provider": "llamacpp",
      "maxTokens": 4096
    },
    "COMPLEX": {
      "model": "gemini-2.5-pro",
      "provider": "google",
      "maxTokens": 8192
    },
    "FRONTIER": {
      "model": "claude-opus-4",
      "provider": "anthropic",
      "maxTokens": 16384
    }
  },
  "classifier": {
    "aiThreshold": 0.7,
    "aiTimeout": 3000,
    "aiModel": "gemma-4-e2b"
  },
  "escalation": {
    "enabled": true,
    "maxRetries": 2,
    "retryDelay": 1000
  }
}

.env.example

# ─── API Keys (at least one required) ───────────
OPENAI_API_KEY=
ANTHROPIC_API_KEY=
GOOGLE_API_KEY=
DEEPSEEK_API_KEY=

# ─── Server ─────────────────────────────────────
LAROUTER_PORT=18790
LAROUTER_HOST=127.0.0.1
LAROUTER_ADMIN_KEY=

# ─── Local Models ───────────────────────────────
LLAMA_CPP_PATH=/usr/local/bin
LAROUTER_OLLAMA_URL=http://localhost:11434

# ─── Logging ────────────────────────────────────
LAROUTER_LOG_LEVEL=info
LAROUTER_DB_PATH=./larouter.db

Runtime Configuration API

Update configuration without restarting:

# View current config (keys redacted)
curl http://127.0.0.1:18790/api/config

# Update tier mapping
curl -X PATCH http://127.0.0.1:18790/api/config \
  -H 'Content-Type: application/json' \
  -d '{"tiers": {"SIMPLE": {"model": "gemma-4-26b-a4b"}}}'

Per-Project Overrides

Each project can override the global tier mapping:

{
  "name": "my-app",
  "routingPolicy": "cost-optimized",
  "tierOverrides": {
    "COMPLEX": {
      "model": "gemini-2.5-flash",
      "provider": "google"
    }
  },
  "budgetUsd": 50.0
}

Available routing policies:

Policy	Behavior
balanced	Default — use tiers as configured
cost-optimized	Prefer local models, only escalate when necessary
quality-optimized	Prefer cloud models for all non-trivial tasks
local-only	Never use cloud APIs