Skip to content

configuration.md

Latest commit

 

History

History
124 lines (93 loc) · 3.48 KB

configuration.md

File metadata and controls

124 lines (93 loc) · 3.48 KB

Configuration

Server Configuration

SourceBridge.ai reads configuration from environment variables and config files.

Environment Variables

Variable Description Default
SOURCEBRIDGE_PORT API server port 8080
SOURCEBRIDGE_HOST API server host 0.0.0.0
SOURCEBRIDGE_JWT_SECRET JWT signing secret (required for auth)
SOURCEBRIDGE_LOG_LEVEL Log level (debug, info, warn, error) info
SOURCEBRIDGE_TEST_MODE Enable test mode (no auth required) false

Config File

Default location: ~/.sourcebridge/config.yaml

server:
  port: 8080
  host: 0.0.0.0

auth:
  jwt_secret: your-secret-here
  token_expiry: 24h

logging:
  level: info
  format: json

LLM Configuration

Code reasoning features (review, discussion, explanation) require an LLM provider.

Supported LLM Providers

Configure via SOURCEBRIDGE_WORKER_LLM_PROVIDER or [worker] llm_provider in config.toml. The worker validates the value at startup and refuses to start with an unknown provider, printing an actionable error that names the supported set.

Provider Config value
Anthropic anthropic
OpenAI openai
Ollama (local) ollama
vLLM vllm
llama.cpp llama-cpp
SGLang sglang
Google Gemini gemini
OpenRouter openrouter
LM Studio lmstudio

Supported Embedding Providers

Embeddings are configured independently from the LLM provider. Use SOURCEBRIDGE_WORKER_EMBEDDING_PROVIDER or [worker.embedding] provider in config.toml. As with LLM providers, an unknown value is rejected at startup.

Provider Config value Notes
Ollama ollama Default. Requires a model like nomic-embed-text.
OpenAI openai OpenAI hosted embeddings (text-embedding-3-*).
OpenAI-compatible openai-compatible Any self-hosted endpoint with the OpenAI embeddings API shape.

Note: anthropic is not a valid embedding provider — Anthropic does not offer an embeddings API. Setting it produces an error at startup that names this explicitly.

Embedding env vars use the SOURCEBRIDGE_WORKER_EMBEDDING_ prefix:

SOURCEBRIDGE_WORKER_EMBEDDING_PROVIDER=ollama
SOURCEBRIDGE_WORKER_EMBEDDING_MODEL=nomic-embed-text
SOURCEBRIDGE_WORKER_EMBEDDING_BASE_URL=http://localhost:11434
SOURCEBRIDGE_WORKER_EMBEDDING_DIMENSION=768

See config.toml.example for the full [worker.embedding] section with comments.

Test Mode

When SOURCEBRIDGE_TEST_MODE=true, the system uses a FakeLLMProvider that returns deterministic responses. This is used for testing and CI.

Docker Compose Configuration

The docker-compose.yml file configures all services:

services:
  api:
    image: ghcr.io/sourcebridge/sourcebridge:latest
    ports:
      - "8080:8080"
    environment:
      - SOURCEBRIDGE_JWT_SECRET=change-me
      - SOURCEBRIDGE_LOG_LEVEL=info

  web:
    image: ghcr.io/sourcebridge/sourcebridge-web:latest
    ports:
      - "3000:3000"
    environment:
      - NEXT_PUBLIC_API_URL=http://api:8080

  worker:
    image: ghcr.io/sourcebridge/sourcebridge-worker:latest
    environment:
      - OPENAI_API_KEY=${OPENAI_API_KEY}

VS Code Extension Configuration

In VS Code settings:

Setting Description Default
sourcebridge.apiUrl API server URL http://localhost:8080
sourcebridge.token JWT token for auth (empty)