Settings keep their GRAFT_ names. See the repo’s .env.example for the full list (model, base URL, graph directory).
| Variable |
Description |
GRAFT_PROVIDER |
openai for any OpenAI-compatible endpoint, anthropic for the native API, or litellm / orcarouter for a gateway that speaks the OpenAI-compatible format |
GRAFT_API_KEY |
Your provider key |
GRAFT_MODEL |
The model to use |
GRAFT_BASE_URL |
For the openai wire format: point at OpenRouter, Fireworks, Groq, a LiteLLM proxy, a local server, or OpenAI itself |
OPENROUTER_API_KEY |
Deprecated fallback for the key |
ORCAROUTER_API_KEY |
A second fallback for the key |
The same settings can be passed as --provider/--model/--api-key/--base-url on the command line.
| Variable |
Description |
GRAFT_NO_REFRESH=1 |
Answer from the graph exactly as it is on disk, for every command (same as --no-refresh) |
GRAFT_REFRESH=hash |
Hash every file instead of trusting size+mtime |
| Variable |
Description |
GRAFT_NO_STATUSLINE=1 |
Skip writing the Claude Code statusLine (same as trail init --no-statusline) |
| Variable |
Description |
DO_NOT_TRACK=1 |
Disables both network calls: the daily npm version check and the usage ping |