.env.example

# Environment variables for Tolokaforge
# Copy this file to .env and fill in your API keys:
#   cp .env.example .env

# At least one LLM provider key is required.
# Most examples use OpenRouter by default.
OPENROUTER_API_KEY=your-openrouter-api-key-here

# Optional: direct provider keys (used when provider is set to "anthropic", "openai", etc.)
# ANTHROPIC_API_KEY=your-anthropic-api-key-here
# OPENAI_API_KEY=your-openai-api-key-here
# GOOGLE_API_KEY=your-google-api-key-here

# Gemini direct (Google AI Studio): use when provider=gemini in ModelConfig.
# Create at https://aistudio.google.com (no GCP IAM required; key is a bearer token).
# Bypasses OpenRouter — useful for discriminating model behavior from provider transport.
# GEMINI_API_KEY=your-gemini-studio-api-key-here

# Vertex AI (GCP): use when provider=vertex_ai in ModelConfig.
# Service account needs role roles/aiplatform.user (or aiplatform.endpoints.predict).
# Download a JSON key, point GOOGLE_APPLICATION_CREDENTIALS at it.
# GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json
# VERTEXAI_PROJECT=your-gcp-project-id
# VERTEXAI_LOCATION=us-central1

# LLM gateway (a LiteLLM proxy or a gateway presenting the same surface).
# Set the base URL to route calls through the gateway instead of calling
# providers directly; model names and configs stay unchanged.
# See docs/LLM_LAYER.md § proxy.
# LLM_PROXY_BASE_URL=https://gateway.example.com
# LLM_PROXY_API_KEY=your-gateway-key-here
# Static attribution headers your gateway requires, as a JSON object:
# LLM_PROXY_HEADERS={"X-Team-Id":"research","X-Cost-Center":"42"}
# A value may reference a secret as ${secret:NAME}, so this string stays non-secret
# even when a header carries something sensitive. See docs/LLM_LAYER.md.
# LLM_PROXY_HEADERS={"X-Team-Id":"research","X-Order-Id":"${secret:ORDER_ID}"}
# ORDER_ID=...
# Header name to fill with a fresh UUID per request:
# LLM_PROXY_REQUEST_ID_HEADER=X-Request-Id
# Which providers to route. Default is openrouter,openai — the only ones whose
# litellm transport sends an OpenAI-envelope request. Widen this only if your
# gateway also serves the other provider's native route (see docs).
# LLM_PROXY_PROVIDERS=openrouter,openai
# Namespace that wins when the gateway serves one model under several names:
# LLM_PROXY_PREFERRED_ROUTE=openrouter/