Changelog

🎙️ Auto Transcribe with Granola, Auto Models & Bring Your Own Models

Never take meeting notes again (😱) with Auto Transcribe, powered by Granola. Your meetings can now be transcribed and usefully summarized so you can stay focused on the discussion. Auto Transcribe works together with Auto Join Meetings so you can seamlessly jump into calls while Granola captures all the important details in the background.

Go to Settings → Extensions → My Schedule to enable and customize. You can decide which meetings this applies to as well as whether to show a confirmation before joining so you’re never caught off guard before joining your next call.

✨ New

• AI Chat Text Settings: You can now control the text size and line spacing in AI Chat independently of the main window. See Settings → AI → AI Chat

🧪 New AI Experiments

• Auto Model: Let AI choose the best model for the job at hand. We’ve enabled this by default so you can simply pick Auto when choosing a model and the most suitable model will be used for your prompt. We’re currently testing this (hence, experimental!) and will be fine tuning it as we receive your feedback. Please use the 👍 👎 and let us know how it’s going in #ai-experiments. Check out the AI Manual for more details.
• Bring Your Own Models: For advanced users you can now add any OpenAI compatible LLM provider to Raycast AI. BYOM (Custom Providers) is disabled by default.

Note: To enable or disable AI Experiments go to Settings → AI → Experiments

🐞 Fixes

• Root Search: Do not present AI Extensions popover for @ prefix
• AI Chat: Fixed chat image capture layout
• AI Chat: Ensure AI Chat window is not shown before screen capture
• AI Chat: Ensure Escape cancels screen area capture
• AI Chat: Starting a new chat no longer cancels current completion if chat branching is enabled
• Markdown: Fixed hang when processing HTML comments

🌿 Chat Branching (Experimental)

Create alternate conversation paths from any point in your chat history. Think of it as a "save point" where you can explore different directions without losing your original conversation. With background execution, you can kick off multiple branches that continue processing even when you switch between chats.

To use: From any AI Chat, simply press CMD + Shift + B to create a new branched chat and continue the conversation. Alternatively, you can right click on a chat in the sidebar, and select Branch Chat.

Note: This is an experimental feature. We’d like to see if you find it useful and how we might improve it. We’ve enabled it by default as it won’t disrupt your existing workflow, but if you’d like to disable it, go to AI Settings → Experiments. Please share your thoughts and feedback in #ai-experiments.

✨ New

• OpenRouter: We’ve added support for OpenRouter in order to give you even more control and flexibility over your AI usage in Raycast. OpenRouter provides access to a wide range of public models and has a beautiful, easy interface to manage your usage and costs without a subscription. To use: Add your OpenRouter API key in AI Settings → Custom Providers → OpenRouter. Once the key is validated, OpenRouter models will appear in the model picker.
• Clipboard History: Added a new Paste Sequentially command allows you to paste previously copied content one after another into your frontmost application. It’s best used with a hotkey, e.g. ⌥ + ⇧ + V.
• Calendar: Added support Riverside and StreamYard as conference providers for upcoming meetings.

💎 Improvements

• AI: Added a Show Tool Call Info switch in Settings → AI. When enabled you can view arguments and output for all tool calls in the chat
• AI: We now show a warning in the chat if there is an on-going incident with the LLM provider
• AI Chat: Composer text is now preserved when switching between chats
• AI Chat: Improved handling of presets in new chat menu
• MCP: The server version is now displayed when viewing a server in Manage Servers
• MCP: Added DENO_PATH and NODE_PATH to stdio environment

🐞 Fixes

• AI: Always use the original image when saving and sharing a chat image
• AI: Fixed cursor jumping after mentioning an AI Extension in root search
• AI: Fixed chat message layout corruption in Tahoe developer betas
• AI: Improved model fallback handling when using an API key
• AI Chat: Fixed jumping to first pinned chat after completion when >20 pinned items
• AI Chat: Generated chat titles should now use the user’s language
• Quick AI: Fixed Regenerate with Model action for commands
• MCP: Fixed the Streamable HTTP client accept header
• MCP: Fixed an issue parsing large numbers
• MCP: The HTTP client is now compatible with Atlassian’s MCP server

🔑 Bring Your Own Key

Bring Your Own Key (BYOK) seamlessly integrates your existing AI provider accounts with Raycast's intuitive interface. With BYOK you can now use Raycast AI with your own API keys for Anthropic, Google, and OpenAI. This allows you to send as many AI messages as you want at your own cost without a Pro subscription. To add an API key, open Settings -> AI and scroll down to the Custom API Keys section.

✨ New

• Manage Models: A new command to view and manage all your AI models. You can easily disable individual models or all models from the same provider using the command.
• Experimental AI features: The AI landscape moves quick — what works with one provider today might behave differently tomorrow, or might not be consistently supported across different models. You can now toggle on experimental AI features to try out early-stage features that we’re exploring, testing and trialing. Turn them on in AI Settings and join #ai-experiments in our Slack Community to share your thoughts!
  Available Experiments:
  • 🎏 MCP HTTP Servers: Support for HTTP MCP servers using the SSE (Server Sent Events) and Streamable protocols.
  • 🦙 AI Extensions for Ollama models: If you’ve got Ollama installed, you can try out tool calling with local models. Since tool choice and streaming for tool calls aren’t supported by Ollama just yet, this can be a bit unreliable - which is why it’s experimental 😉

💎 Improvements

• Spotlight: Added new “Replace Spotlight with Raycast” command for a smoother transition to your favorite launcher
• AI Chat: Added new Send Active Window to AI Chat, Send Selected Area to AI Chat, and Send Focused Browser Tab to AI Chat commands to capture context even quicker

🐞 Fixes

• Safari Extension: Fixed an issue where the connection to the app would be closed and couldn’t reconnect. Make sure to update Safari Extension to the latest version in App Store.

🦙 Local Models

Local models allow you to run nearly any open source LLM locally, on your machine. Through our new integration with Ollama, you’ll now have access to more than 100 AI models from various providers ranging from small 135M to massive 671B parameter models.

We’ve also added experimental support for AI Extensions with Local Models. Since tool choice and streaming for tool calls aren’t supported by Ollama just yet, AI Extensions can be a bit unreliable when using Local Models. However, if you want to try it out, you can enable it by going to AI Settings. Keep in mind, it likely won’t be as reliable as using non-local models.

✨ New

• Local Models: Get started by installing Ollama. Then download Ollama models directly from the Raycast Settings, in the Local Models section of the AI tab by copy & paste model names. You can find the list of all available Ollama models here.
• Local Models: Support for Vision with local models supporting it. You can find the list of supporting models here.
• Local Models: Experimental support for AI Extensions with local models that support tool calls. You can find the list of supporting models here.

💎 Improvements

• MCP: Improved error reporting when stdio servers fail to run
• MCP: Improved compatibility with server JSON schemas
• MCP: Added a Copy to Clipboard action in Manage Servers

🐞 Fixes

• Onboarding: Fixed image assets not loading in the onboarding pages
• AI: Fixed issue where community AI Extensions were not disabled when AI was disabled
• AI: Fixed remote tool calls in AI Commands
• AI: Disable default tools in AI Commands
• AI: The AI Chat/Quick AI Web Search setting now works even if the Ask Web command is disabled
• MCP: Fix handling of quoted server arguments
• MCP: Server updates are no longer saved if the updated server fails to run
• Export: Prevent the export/import view from resizing the main window

🧰 MCP

Raycast now includes integration with the Model Context Protocol (MCP). You can install any local stdio server and then access the server’s tools from Quick AI and AI Chat by @-mentioning them, just like with AI Extensions.

Learn more about MCP in our manual and discover MCP servers with our new Registry extension.

✨ New

• MCP: New commands to install, import and manage servers
• MCP: New Registry to explore servers available as a separate extension

💎 Improvements

• AI: Added a Confirm Always action for AI Extension / MCP Server tool confirmations. You can reset allowed tools in the AI Extension / MCP Server settings or in the general AI settings
• AI Commands: The @location tool must now be explicitly mentioned to be used from commands

🐞 Fixes

• AI: Fixed a crash happening when opening a markdown link while the message was rendered
• AI: Fixed a rare crash happening when pasting file attachments to AI Chat
• AI: Fixed an occasional crash that occurred when using the @location AI Extension

📱 Raycast for iOS is here

The wait is over! Raycast for iOS is finally here after being one of our most requested features. The app brings the power of your desktop experience to mobile:

• 💬 Chat with dozens of AI models for perfect answers on any task
• 📚 Capture and organize thoughts so brilliant ideas never slip away
• 🔗 Access your Quicklinks instantly for important resources on the go
• 📝 Use your code snippets and templates to maintain productivity anywhere
• ☁️ Sync seamlessly across all your Apple devices with zero friction

Wanna learn more? Watch this walkthrough with Pedro to get the most out of the app or read this blog post about how we built it.

✨ New

• AI Extensions: You can now use our AI Extensions with a variety of different models such as GPT-4.1, Claude 3.7 Sonnet, Gemini 2.5 Flash, and more. This way, you can @-mention your apps and services in your chats to gather context or perform actions. Try it out with @finder to save your latest research as a CSV table or @calendar to timebox your priorities for next week.
• AI: Recently used models remain ranked higher across the app’s model picker

💎 Improvements

• AI: Added icon for reasoning effort next to model name in AI Chat

🐞 Fixes

• AI: Fixed an issue where the temperature was shown in AI Chat for models that don’t supported it
• AI: Fixed an issue preventing AI from being disabled caused by ad-hoc AI Commands
• AI: Fixed AI messages left count displayed in preferences window when AI is not enabled
• AI Chat: Fixed incorrect action bar button text color

❇️ Try Raycast AI For Free

We’ve just made Raycast AI available to everyone. You get 50 free messages to use across AI Chat, Quick AI, and AI Commands - no subscription or account needed. Experience leading models with powerful extensions directly on your Mac!

Here’s what you can do with Raycast AI:

• Quick answers: Type a question and press Tab to get instant answers with web citations
• Text improvements: Select text in any app and use AI Commands like "Fix Spelling and Grammar" to correct typos
• Dozens of models: From OpenAI’s o3-mini to Google’s Gemini 2.0 Flash, DeepSeek’s V3, and beyond. Select and switch to the ideal model for any given task.
• Context-aware assistance: Attach PDFs, browser tabs, or screenshots to get help with what you're working on.
• Extension integration: Interact and instruct using natural language with AI Extensions - for example, you can type @focus to get in the flow on your top @linear issues

Want to learn more? Watch a walkthrough video or check out our new AI manual for all the details.

Not into AI? No problem! You can easily turn it off via Raycast settings → AI tab → big toggle on the left.

✨ New

• New Models: Added new LLMs to Raycast AI throughout the last few days. Here’s a quick overview:
  • OpenAI’s GPT-4.1, GPT-4.1 mini and GPT-4.1 nano
  • Google Gemini 2.5 Pro
  • Meta Llama 4 Scout powered by Groq
  • DeepSeek v3 powered by Together
• Screenshots: Added a Storage Duration preference. Setting a shorter duration will automatically move older screenshot files to the Trash once a day. The default is Unlimited.
• Screenshots: Added a preference for the primary action: Copy to Clipboard or Paste to Active App

💎 Improvements

• The “Send Feedback” command now opens a separate window and accepts file uploads

🐞 Fixes

• Fixed delay showing Calculator results for operators that matched Flight designators (e.g. bin 3 or hex 50).

✨ Improvements and bug fixes

Some releases are bigger than others. This time we’ve been working away on improvements and bug fixes to enhance your Raycast experience - enjoy!

✨ New

• Reasoning Effort: You can now set the reasoning level for OpenAI’s o3-mini and Claude 3.7 Sonnet (Reasoning). Increase it when you need deep analysis or dial it down for quicker, more straightforward problems. You can change it in the settings of your AI Chat or Presets. It isn’t supported in AI Commands yet.
• Favicon Provider: We added a new setting to let you pick your favicon provider or disable it altogether, allowing you to balance user experience and privacy. You can configure it via Raycast Settings → Advanced → Favicon Provider. Read more about it here.
• macOS Support: Raycast now requires macOS 13 or later. If you are still running macOS 12 you should consider upgrading to ensure you continue to receive updates to Raycast.

💎 Improvements

• Clipboard History: Changed the visual information for links setting to also enable or disable favicons
• AI: Improved handling AI Extension @mentions in root search
• AI Chat: Significantly improved performance when loading large code blocks
• AI Chat: Improved scroll to bottom behaviour during completion
• AI Chat: It should now be possible to copy code blocks during completion
• AI Chat: Improved responsiveness during live resize
• AI Chat: Enabled LaTeX in user messages
• Clipboard: Passwords should now be ignored when copied using the Passwords menu bar extension. Note: Requires the Passwords app to be excluded from Clipboard History and requires Accessibility permissions for Raycast.
• Calculator: The text in the search bar won’t be selected when performing calculations and temporarily hiding and reopening the Raycast window, allowing you to easily append text to the expression.
• Raycast Notes: Added an action to duplicate notes

🐞 Fixes

• AI: LaTeX $$ delimiters are no longer recognized within code blocks
• AI: Improved handling of LaTeX \begin \end commands
• AI: Fixed an issue where the selected text would be cleared during completion
• AI Chat: Fixed layout of chat settings panel after changing model
• AI Chat: Fixed handling of images when receiving multiple responses from the LLM
• AI Chat: Fixed issues with inline code block layout
• Quit All Apps Except Frontmost: Improved handling when Finder or Raycast are frontmost apps
• Focus: Improved Firefox based browsers website blocking reliability

⏲️ Auto-Join Meetings

Do you ever manage to get a lot of work done just before a meeting? We don’t either. It’s hard to concentrate when you’re worried about missing an upcoming meeting. But if we’re reminded at the right time… we can focus, even if it’s just for an extra 30 minutes.

This is why we’re releasing Auto-Join Meetings. When enabled and a meeting is about to begin, Raycast will show a prompt to join the meeting. If your conferencing app has its own confirmation screen already or you feel extra spicy, you can disable the alert and let Raycast auto-join you to meetings right away. You’ll only get auto-joined to events in your enabled calendars which have a meeting link attached.

Go to the My Schedule command preferences to enable it and give it a shot!

✨ New

• Workspace: Added the “Quit All Apps Except Frontmost” command. Use this feature to focus on a single task by quitting all other applications.

💎 Improvements

• AI Chat: Improved overall performance with long messages and attachments
• AI Chat: Added support for sending empty messages with attachments
• Focus: Added a toggle in the extension setting to allow all websites in Allow mode if a browser is allowed but no website is specified. Raycast Settings > Extensions Tab > Raycast Focus > Allow all Websites.

🐞 Fixes

• AI Chat: Fixed an issue where 1…9 indicators would get stuck in the list sometimes when switching apps
• Organizations: Fixed an issue with the organization upgrade button opening an incorrect URL
• Focus: Fixed website blocking not working with renamed browser apps

AI Extensions Beta

Part of what makes Raycast special is our rich, customizable extension ecosystem. As of today, extensions are getting a little smarter, and a lot more useful.

Introducing: AI Extensions

Setup a coding workflow, organize your day, book some time with a colleague on the fly... AI Extensions let you do this directly from Raycast without ever having to open an app or the browser. And all of this using simple, natural language.

Find your favorite extensions, now AI enabled, using the new filter in the Store. Once installed, simply type @ to explore the possibilities. AI Extensions can be used across Raycast AI. They’re supported in Quick AI, AI Chat, AI Commands, and AI Chat Presets.

Learn more about the new feature here.

Learn how to use AI Extensions from Thomas and Pedro → https://youtu.be/sHIlFKKaq0A?si=RiGamvGhg3Hkrk2z

💎 Improvements

• Raycast Focus: Added an option to define apps & websites you want to allow instead of blocking them. In Allow mode, all apps and websites except the ones you enter will be blocked. Find the switch in the Start Focus Session command.
• Raycast Focus: Added the Allow mode to the Shortcuts and system Focus filter. You may need to restart your Mac for the updated Shortcuts and filter to work.

🐞 Fixes

• Raycast Focus: Fixed label contrast in indefinite session menu bar item
• Open In Editor: Fixed an issue where opening items with the default editor would incorrectly open Finder instead