Skip to content
AI

Chat

Raycast offers two ways to work with an AI model. Quick AI handles fast, one-off questions straight from Root Search, while AI Chat is a full multi-turn conversation interface with persistent history, Agents, memory, attachments, and tool use. Together they cover everything from a quick lookup to a longer working session, all without leaving Raycast.

Configure the AI models used across Quick AI, AI Chat, and Custom AI Commands from Raycast Settings → AI → Models, or use the Manage Models command to jump straight there.

Pick the default model for each surface:

  • Quick AI: the model used when you ask a quick question from Root Search.
  • AI Chat: the default for new chats. Set to Last Used Model to continue with whichever model you used most recently.
  • Custom AI Commands: the default for any AI Command that doesn't specify its own model.

Choose which models appear in the model pickers across Quick AI, AI Chat, and Custom AI Commands:

  • Review model details: each model lists its Speed, Intelligence, and Context window so you can compare capabilities at a glance.
  • Enable or disable models: toggle the checkbox next to a model to control whether it shows up in pickers. Disabled models are hidden everywhere AI models can be selected.
Toggling a model on or off in Manage Models
  • Group by Provider: models are grouped by their provider (e.g. Raycast, Anthropic). Expand or collapse a provider to focus on a subset.
  • Sort the list: use the sort menu in the top-right to order models by Brand, Alphabetically, Speed, Intelligence, or Context Window.
Sort menu in Manage Models

Quick AI is the fastest way to ask a one-off question from Raycast, without leaving your flow. Start from Root Search, hit Tab, and get an answer in the same window.

New Quick AI now uses a chat-style interface with follow-ups, context attachments, and a one-keystroke handoff to AI Chat.

Quick AI in Raycast

There are three ways to start:

  • From Root Search: start typing your question, then press Tab. Raycast hands your text to Quick AI and submits it.
  • Run the Quick AI command directly from Root Search to open with an empty prompt.
  • Set Quick AI as a fallback command in Raycast Settings → Launcher → Fallback Commands to send any unmatched Root Search text straight to Quick AI when you press .

Pressing Tab from Root Search always opens Quick AI, even with no text typed, so you can start fresh.

Type your prompt and press to submit. Quick AI streams the answer back in the same window.

On the response, pastes the answer into the previously focused app by default. To swap the default to copy instead, open Raycast Settings → AI → Commands → Quick AI → Primary Action.

When a response includes code, it appears in a code block with the detected language and a copy button. Long lines scroll sideways by default. Click Enable line wrap in the top right of the block to wrap them to the response width, and click again to switch back.

Enable line wrap button in a Quick AI code block

Keep typing after a response to ask a follow-up. Quick AI is a full conversation, not a single-shot Q&A.

  • Regenerate the last response with the same model: / Ctrl R
  • Regenerate the last response with a different model: / Ctrl R
  • Start a fresh conversation: / Ctrl N
  • Navigate between recent Quick AI chats: / Ctrl [ and / Ctrl ]

Open the Action Panel ( / Ctrl K) and pick Change Model to switch the active model mid-conversation, or Regenerate with Model… to try the last answer on a different model.

You can also set a default in Raycast Settings → AI → Commands → Quick AI → Default Model.

When a question grows into something bigger, press / Ctrl J to move the conversation into AI Chat. You keep the full history and get the chat workspace, longer context, and tools.

Quick AI starts a fresh chat after a period of inactivity, so you don't accidentally tack new questions onto an older conversation. Adjust the timeout in Raycast Settings → AI → General → Start New Chat (options: 5 minutes, 10 minutes, 15 minutes, 30 minutes, 1 hour, always, never). This same inactivity timer also determines whether a Send to AI command reuses your active chat or starts a new one when AI Chat is closed.

Tune Quick AI from Raycast Settings → AI → Commands → Quick AI:

  • Quick AI Default Model: default model used when opening Quick AI.
  • Primary Action: paste to active app or copy the response to clipboard on .
  • Tab Shortcut: hide the Tab hint in Root Search (the shortcut still works).

AI Chat is Raycast's full chat workspace. Start a multi-turn conversation, attach context from your screen, give the model tools, and pick up where you left off across days.

New AI Chat is rebuilt in Raycast 2 with persistent history, Memory, rich attachments, and tool use, a full chat experience that replaces the single-shot AI flows in classic Raycast.

AI Chat workspace in Raycast

There are a few ways to start:

  • Run the AI Chat command from Root Search to open the chat workspace.
  • Press / Ctrl J from a Quick AI conversation to continue it in AI Chat with the full history.
  • Use a Send to AI command (Send Focused Window, Screen Area, Screen, or Selected Text) to send content into your active AI Chat from anywhere.

AI Chat lives in its own window, so you can keep it alongside whatever you're working on.

Type your prompt in the composer and press to send. adds a new line. AI Chat streams the response back and remembers the whole thread, so each follow-up can build on earlier turns.

  • Edit any previous message to re-run the conversation from that point.
  • Regenerate the last response with / Ctrl R.
  • Start a fresh chat with / Ctrl N.
  • With an empty composer, / step through your recent messages.

When a response includes code, it appears in a code block with the detected language and a copy button. Long lines scroll sideways by default. Click Enable line wrap in the top right of the block to wrap them to the response width, and click again to switch back.

Enable line wrap button in an AI Chat code block

Open the model picker from the composer. Pick a fast model for chat, a reasoning model for harder problems, or an image-capable model when you want to generate visuals.

Tune model behavior from the settings sidebar:

  • Creativity: None, Low, Medium, High, or Maximum.
  • Reasoning Effort: for models that support extended thinking.
  • Web Search: let the model pull live results during the answer.

Turn on Memory (in Raycast Settings → AI → Personalization → Memory) and AI Chat will remember durable facts across conversations: your name, the projects you work on, the format you prefer answers in. The model writes to and reads from memory automatically as you chat.

Hit the attach button in the composer or type @ to add context to the next message. AI Chat can pull from:

  • Files: documents, code, images.
  • Notes: entries from Raycast Notes.
  • Clipboard History: any recent clipboard item.
  • Focused Window, Window…, Selected Area, or Entire Screen: screenshots from your desktop.
  • Browser Tabs: open tabs from supported browsers.
  • Calendar Events: today or upcoming.
  • Web Search: live results during the answer.

Attachments are scoped to the message you send them with, but the model can keep referring back to them later in the chat.

When you use a Send to AI command from elsewhere in Raycast — Send Focused Window to AI, Send Screen Area to AI, Send Screen to AI, or Send Selected Text to AI — the content goes to your currently active chat instead of always starting a new one. If AI Chat is already open, it's sent straight to the active chat. If AI Chat is closed, your Start new chat after inactivity setting (see Auto-new Chat) decides what happens: within the window the active chat is reused, and once it's elapsed a new chat begins. The one exception is streaming — if you close AI Chat while the active chat is still generating a response, that chat is always restored on next open, regardless of the timer.

AI Chat can call tools: run a web search, generate an image, execute a terminal command, invoke an AI Extension, or hand work off to an MCP-compatible server. Toggle which tools are available per chat from the Tools section of the settings sidebar. With Tool Confirmation on, AI Chat asks before each tool runs.

Switch the chat to a saved Agent with the Change Agent action in the Action Panel. Agents set instructions, model, and AI Extensions for consistent results until you switch agennts or choose Ask Anything. See the Agents page page on creating your own.

Author reusable knowledge for the model with plain SKILL.md files. Raycast discovers them automatically and loads them when they're relevant to your chat. See the Skills page for the full format and where to put them.

The history sidebar shows every chat, with search at the top. From the Action Panel ( / Ctrl K) on any chat you can:

  • Pin important conversations to the top.
  • Archive chats you're done with. Old chats archive automatically after the window you pick in Settings (or never).
  • Delete for good.

Toggle Always on Top from the Action Panel to keep AI Chat hovering over your other apps, handy while you're chatting about something on screen. You can also set this up in Raycast Settings → AI → Commands → AI Chat → Always on Top.

When a Quick AI question grows into something bigger, press / Ctrl J to move it into AI Chat. The full history, model, and any attachments come with it.

Tune AI Chat from Raycast Settings → AI:

  • Start New Chat: start a new chat after the specified timeout when opening Quick AI, and determines whether a Send to AI command reuses the active chat or starts a new one when AI Chat is closed.
  • Send Message: keyboard shortcut used to send messages in AI Chat ( or / Ctrl ).
  • Conversation History: show conversations from AI Chat and Quick AI combined or keep them separate.
  • Auto-Archive Chats: automatically archive chats after the specified period of inactivity.
  • Always on Top (per-command, under AI Chat): keep the AI Chat window above other windows.

AI Chat and Quick AI let you talk to models from anywhere in Raycast. Most reports come down to the trigger key, model selection, sync, or hitting a usage or context limit. Here are the common ones.