Changelog

Extensions can now provide their own AI models to Raycast. Just install a compatible extension from the Store, complete its setup, and choose its models in AI Chat. It’s an easy way to bring even more models into your everyday workflow.

We updated a few extensions: Ollama AI, AI by Vercel, Cerebras, Groq, LM Studio, etc.. Developers, check out the API changelog for more information.

💎 Improvements

• AI Chat: Rename a chat by double-clicking its title in the top bar. Press Enter to save or Esc to cancel, and the window can still be dragged by the title
• AI Chat: Dictation with Ctrl + M in the AI Chat and Quick AI composers now uses the same processed transcript as the Dictate command. The text shows up once you stop recording instead of streaming into the editor while you speak
• File Search: Added a setting to entirely disable indexing of network/removable/virtual volumes
• File Search: Indexing now respects the .metadata_never_index marker file convention from Spotlight
• Script Commands: Added line wrapping to full output so long lines are easier to read

🐞 Fixes

• AI: Fixed the tool approval bar not responding in later turns of the same response when using Ollama or OpenAI-compatible custom providers
• AI: Fixed failing requests retrying in the background indefinitely during provider outages. Raycast now stops after a few attempts and shows an error with options to change the model, resume, or stop.
• AI Commands: Disabled automatic tool discovery and skill loading
• File Search: Classify NFS, VirtualBox and other mounted volumes correctly as non-local volumes
• Quick AI: Fixed ctrl + Enter (Copy Response) and alt + Enter (Paste Formatted Response) not working right after a response in some cases
• Store: Fixed Install Extension Anyway failing for extensions marked as unsupported on your platform
• AI Profile: Fixed clearing and saving an empty profile
• Open With…: Fixed the menu initially selecting an unrelated app before recommended apps finished loading
• Calculator: Fixed currency conversions using missing or outdated rates during refreshes