Direct mode
Flames sends chat requests straight to an OpenAI-compatible endpoint. The ReAct loop, tool dispatch and conversation state run on the phone.
- Streaming requests to
/v1/chat/completions - Hermes Server, OpenAI, Anthropic, Gemini, Groq, OpenRouter, 9router or a local Ollama
- Photos and files can be attached to a turn
- Memory search runs against a SQLite FTS5 index on the device