Using Levix
AI providers
Gemini, OpenAI, Anthropic, or anything OpenAI-compatible — picked in the panel.
The AI agent behind !gemini isn't tied to one vendor. Settings → AI has a provider registry with a preset for each of these — pick one, paste a key (or none, for a local model), and Levix takes it from there:
| Provider | Notes |
|---|---|
| Google Gemini | The default. Also the only provider !generate (image generation) and one of the two speech-to-text options run on. |
| OpenAI | Chat, vision, and Whisper transcription. |
| Anthropic | Claude models, chat and vision. |
| Groq Cloud | An OpenAI-compatible endpoint — fast, hosted open models. |
| Ollama (local) | Point the base URL at your own machine — a fully local model, no API key at all. |
| Any other OpenAI-compatible endpoint | OpenRouter, LM Studio, or anything else that speaks the OpenAI chat-completions API — set a custom base URL. |
Model discovery, not a hardcoded list
Once a key is set, Levix calls the provider's own models endpoint and shows what your key can actually use — never a model your account can't reach. Each model carries capability flags: vision, speech-to-text, PDF input and more.
Vision and speech-to-text turn themselves off
If the model you pick doesn't support vision or speech-to-text, those toggles in the panel grey out and switch off on their own — there's no state where a feature looks enabled but silently fails.