AI / LLM Settings
Configure the LLM used for summaries, decoration, and Ask AI chat. Echosy supports both an on-device model (MLX, Apple Silicon only) and any OpenAI-compatible cloud provider. You can pick different models per task if you want — e.g., on-device for private summaries, cloud for long-context chat.
On-Device LLM (MLX)
Built into Echosy on Apple Silicon. Downloads the model once, runs locally, no API key, no network, no rate limit. Available choices:
| Model | Size | Notes |
|---|---|---|
| Gemma 4 E4B (4-bit) | ~3 GB | Default — fast, solid quality |
| Qwen 3.5 4B (4-bit) | ~3 GB | Strong multilingual, especially Chinese / Japanese |
| Gemma 4 E12B (4-bit) | ~8 GB | Higher quality, slower; needs 16 GB RAM |
Pick a model in Settings → AI / LLM and click Download. Models stream in with a progress bar; cancel and resume is supported. Once downloaded the model stays on disk — no re-download across launches. You can switch models at any time.
Cloud Provider
Use any OpenAI-compatible endpoint. Each provider auto-fills the API endpoint:
| Provider | Notes |
|---|---|
| OpenAI | GPT-4o, GPT-4o-mini, etc. |
| Gemini | Google's Gemini models |
| Claude | Anthropic's Claude models |
| Groq | Fast inference for open-source models |
| DeepSeek | DeepSeek models |
| OpenRouter | Multi-provider gateway, access many models |
| Ollama | Free, local LLM runner — no API key needed |
| Custom | Any OpenAI-compatible endpoint |
API Key
Your API key for the selected cloud provider. Not needed for on-device or Ollama. The key is stored in your local settings file and is never sent anywhere except to the provider's API endpoint.
API Endpoint
Auto-filled based on your selected provider. You can customize this for self-hosted setups, reverse proxies, or custom OpenAI-compatible servers.
Model
The model name to use (e.g., gpt-4o, gemini-pro, llama3). For on-device, pick from the built-in list. For Ollama, a dropdown shows all locally installed models.
Test Connection
Sends a test request to verify your API key and endpoint work correctly. A success message confirms the connection, or an error message helps you diagnose the issue.
Custom Summary Prompt PRO
Customize how the AI generates summaries. The default prompt produces structured meeting minutes. Examples:
- "Create meeting minutes with action items and decisions"
- "Write a brief executive summary in bullet points"
- "Extract key takeaways and follow-up tasks"
- "Summarize in the same language as the transcript"