Configure LLM Providers
Graft’s --deep mode uses an LLM to generate summaries. You can use any supported provider — your key, your model, your choice.
Supported providers
Section titled “Supported providers”| Provider | GRAFT_PROVIDER value |
Example models |
|---|---|---|
| OpenAI | openai |
gpt-4o, gpt-4o-mini |
| Anthropic | anthropic |
Claude Sonnet, Claude Haiku |
google |
Gemini 1.5 Pro, Gemini Flash | |
| Ollama (local) | ollama |
LLaMA, Mistral, etc. |
| OpenRouter | openrouter |
Any model via OpenRouter |
| LiteLLM proxy | Set GRAFT_BASE_URL |
Any model behind the proxy |
Configuration
Section titled “Configuration”export GRAFT_PROVIDER=openaiexport GRAFT_API_KEY=sk-...export GRAFT_MODEL=gpt-4o-mini # optional, has a defaultexport GRAFT_PROVIDER=anthropicexport GRAFT_API_KEY=sk-ant-...export GRAFT_MODEL=claude-sonnet # optionalexport GRAFT_PROVIDER=googleexport GRAFT_API_KEY=...export GRAFT_MODEL=gemini-1.5-flash # optionalexport GRAFT_PROVIDER=ollamaexport GRAFT_BASE_URL=http://localhost:11434export GRAFT_MODEL=llama3# No API key needed for local modelsexport GRAFT_PROVIDER=openrouterexport GRAFT_API_KEY=sk-or-...export GRAFT_MODEL=anthropic/claude-sonnetSee also
Section titled “See also”- Environment Variables Reference — Complete list of
GRAFT_*variables. - Enable Deep Mode — How to run the deep build.