The six providers
LLM_PROVIDER | Default model | Authentication | Extra install |
|---|---|---|---|
openai (default) | gpt-4.1-mini | GPT3_API_KEY | None |
anthropic | claude-opus-5 | ANTHROPIC_API_KEY, or a stored ant auth login session | pip install anthropic |
gemini | gemini-2.5-flash | GEMINI_API_KEY, or Vertex AI with GOOGLE_CLOUD_PROJECT | pip install google-genai |
ollama | llama3.1 | None | A running Ollama server with the model pulled |
claude-cli | The CLI's own | A signed-in Claude Code CLI | The CLI |
agy-cli | The CLI's own | A signed-in Agy CLI | The CLI |
LLM_MODEL overrides the default for any provider. For openai, the older OPENAI_MODEL setting still works when LLM_MODEL is unset.
Which one keeps my data local?
ollama. It talks to the server at OLLAMA_HOST, http://localhost:11434 by default, so with a local server transcripts never leave the machine. Every other provider sends transcripts to its service.
What if the SDK is missing?
The provider SDKs are imported only when used, so installing one never requires the others. Choosing a provider whose SDK is not installed stops with an error that names the package to install, and an unknown LLM_PROVIDER lists the valid names.