Skip to main content

Configuration

Legible's configuration is split between environment variables (.env) and a YAML configuration file (config.yaml).

Environment Variables​

The primary configuration file is docker/.env. Key variables:

LLM API Keys​

# Google Gemini (recommended)
GEMINI_API_KEY=your-key

# OpenAI
OPENAI_API_KEY=your-key

# Anthropic
ANTHROPIC_API_KEY=your-key

Service Ports​

HOST_PORT=3000                    # Wren UI
AI_SERVICE_FORWARD_PORT=5555     # AI Service
MCP_SERVER_PORT=9000              # MCP Server
MCP_WEB_UI_PORT=9001              # MCP Web UI

Engine Configuration​

WREN_ENGINE_PORT=8080
IBIS_SERVER_PORT=8000
EXPERIMENTAL_ENGINE_RUST_VERSION=v3  # Use Rust engine

AI Service Configuration (config.yaml)​

The docker/config.yaml file configures the AI pipeline in detail.

LLM Provider​

type: llm
provider: litellm_llm
models:
  - model: gemini/gemini-2.5-flash
    kwargs:
      temperature: 0
      max_tokens: 4096

Supported providers and models:

  • Google Gemini: gemini/gemini-2.5-flash, gemini/gemini-2.0-pro
  • OpenAI: gpt-4o, gpt-4o-mini
  • Anthropic: anthropic/claude-sonnet-4-20250514
  • Ollama (local): ollama/llama3, ollama/mistral

Embedder​

type: embedder
provider: openai_embedder
models:
  - model: BAAI/bge-base-en-v1.5
    dimension: 768
api_base: http://text-embeddings-inference:8081
timeout: 120

The default setup uses a self-hosted embedding model (BGE-base-en-v1.5) via the Infinity container. No external API key is needed for embeddings.

Document Store​

type: document_store
provider: qdrant
location: http://qdrant:6333

Engine​

type: engine
provider: wren_ui
endpoint: http://wren-ui:3000

Pipeline Configuration​

Each AI pipeline can be configured with specific model assignments:

type: pipeline
pipes:
  - name: sql_generation
    llm: litellm_llm.gemini/gemini-2.5-flash
    embedder: openai_embedder.BAAI/bge-base-en-v1.5
    engine: wren_ui
    document_store: qdrant

Settings​

settings:
  host: 0.0.0.0
  port: 5555
  column_indexing_batch_size: 50
  table_retrieval_size: 10
  table_column_retrieval_size: 1000
  query_cache_maxsize: 1000000
  query_cache_ttl: 120
  logging_level: INFO

Custom LLM Setup​

To use a different LLM provider:

  1. Edit docker/config.yaml with the new provider configuration
  2. Set the corresponding API key in docker/.env
  3. Restart the AI service:
cd docker
docker compose up -d --force-recreate wren-ai-service

For Ollama (fully local, no API key needed):

type: llm
provider: litellm_llm
models:
  - model: ollama/llama3
    kwargs: {}
api_base: http://host.docker.internal:11434