🚀 New Features
- Voice input on local models is here, so you can talk to the app instead of typing
- Over 20 MCP connectors ship out of the box, including Linear, Notion, Sentry, Atlassian, Stripe, Cloudflare, PostHog, Zapier and Firecrawl
- Agent mode got a proper rework and now runs on MLX and cloud models too
- You can now connect ChatGPT through your Codex subscription
- DeepSeek joins the list of cloud providers
- Ollama and llama.cpp are available as local providers, so you can point Atomic Chat at whatever you already run
- DeepSeek Harness and Muse Code joined the Launch page as well
- Atomic Agent is on the Launch page, a local first agent tuned specifically for local models
🔧 Improvements & Fixes
- Model scans skip GGUF files that carry no weights, so half models stop showing up in your model list
- Model settings, sampling and assistants moved to the top right corner of the chat where you actually reach for them
- 30+ smaller fixes across dialogs, toggles, agents, connectors and the rest of the app
🙏 Contributors
Thanks to @plombeer31, @Vect0rM for their contributions to this release!