The model you want, with your key, or one on the Mac
Gemini, Groq, Mistral, Claude, ChatGPT, Grok, or any OpenAI-compatible API. With LM Studio or Ollama, the model runs on the Mac and needs no key.
Each task can have its own model: Conversation, Voice, Quick tasks, Agents and Screen. A role left empty follows the conversation. When a provider hits a blocking limit, Nemeka switches to the fallback you named until midnight, or for 15 minutes after three failures in a row, and says so in a banner and in the log.
- 9 providers, 2 of them local
- Keys kept in the macOS Keychain
- Tokens and cost of the day per connection









