Configuration
Open source · Apache 2.0. This section documents the open-source core. The Data Workers platform is covered separately - see What the platform adds.
The open-source core is configured by environment variables read at startup. There's no config file to manage and no admin API.
Model provider. Agents call whichever provider you configure, with your key:
| Variable | Provider |
|---|---|
ANTHROPIC_API_KEY | Anthropic |
OPENAI_API_KEY | OpenAI |
AZURE_OPENAI_ENDPOINT, AZURE_OPENAI_API_KEY | Azure OpenAI |
AWS_BEDROCK_REGION | Bedrock |
GOOGLE_CLOUD_PROJECT | Vertex |
OLLAMA_HOST | Ollama (local) |
Transport.
| Variable | Effect |
|---|---|
DW_STDIO=1 | Force stdio transport. This is the default when no health port is set. |
DW_HEALTH_PORT | Run an HTTP health endpoint on this port instead. Defaults to 3000 when enabled. |
Set these in the environment your MCP client launches from - a shell profile or .env.
What you can't configure here
Autonomy levels, approval routing, confidence thresholds and cost guards are properties of a governed deployment, not of a local process. They aren't part of the open-source core, and there are no environment variables that turn them on.
Where the open-source core stops. The open-source agents read, analyse, and recommend. They run against a built-in sample estate so you can drive the full tool surface without wiring anything up. Pointing them at your own infrastructure, letting them act on what they find, and the browser console are the Data Workers platform. → What the platform adds · Book a demo