Changelog

0.48.x

Chat tabs, Custom modes & Faster indexing

This release introduces chat tabs for parallel conversations, a redesigned modes system with custom modes, and improvements to cost visibility, indexing performance, and MCP reliability. Additionally, a sound notification plays when a chat is finished

Built-in modes & custom modes (beta)

Agent and Ask modes are the built-in modes in Cursor, now with the option to add custom modes. We've also renamed "Edit" to "Manual" to better reflect its behavior.

Ask mode now has access to all search tools by default, so the @Codebase tool has been removed. It will automatically search the codebase when needed. If you want to force a search, simply ask Cursor in natural language to "search the codebase". You can disable search from Ask in the mode menu, which will result in Ask only seeing the context you have provided.

Read more about Agent and Ask.

Custom modes (beta) allow you to compose new modes with tools and prompts that fit your workflow. Since custom modes can have custom keybindings, ⌘I will default to Agent mode and ⌘L will toggle the side pane. If you unbind ⌘I, it will also toggle the side pane. Custom modes are currently in beta, and you can enable them from Settings → Features → Chat → Custom modes.

You can also set the default mode from settings (Settings → Features → Chat → Default chat mode) to one of your modes or to the one you used most recently.

Read more about custom modes.

Chat tabs

Create new tabs (⌘T) in chat to have multiple conversations in parallel. You can also hold Option and click the + button to create a new tab. Cmd+N still creates a new chat in the current tab

When a tab is awaiting your input, you'll see an orange dot on that tab.

0:00

-0:00

Faster indexing

We've made significant improvements to indexing performance of similar codebases within a team, greatly reducing the initial indexing time for subsequent copies of large repositories after one copy has been fully indexed. For example, the Cursor codebase now indexes in under a minute, previously taking around 20 minutes.

Sound notification (beta)

Cursor can now play a sound when a chat is ready for review. Enable this feature from Settings → Features → Chat → Play sound on finish.

Usage-based cost visibility

For usage-based models, you can now view the cost and breakdown per chat from the history.

New onboarding

We've introduced an easier onboarding process to help you get started with Cursor. You'll be guided through importing settings, selecting themes, keybindings, and other preferences.

Other (6)
  • Improved indexing performance for teams sharing codebases
  • The vertical sidebar option will temporarily be unavailable for users who opted into Early Access
  • Projects with MCP servers will now prompt users to enable them
  • Chat will now display a notification about usage-based pricing when approaching fast request or usage limits
  • Enhanced visibility of auto-run controls configured by team admins, clarifying when they are active
  • Removed auto-run prompt due to reliability issues. Auto-run has been disabled for all users who previously enabled it
Improvements (4)
  • We've moved "reject all diffs" from Cmd+Backspace (⌘⌫) to Cmd+Shift+Backspace (⌘⇧⌫)
  • MCP on Windows should now be significantly more reliable
  • Improved MCP error messages when configuring MCP servers to assist with debugging
  • Added input token count for chat messages (click the three dots to view). We're continuing to improve context visibility so you can clearly see what gets sent to the model, expect more soon!
Patches (9)
  • 0.48.1 - New onboarding
  • 0.48.2 - Allows users to accept/reject file edited in another chat tab
  • 0.48.3 - Chat Tabs UX improvements (Cmd/Ctrl+T for new tab), Max-mode support for Gemini 2.5 Pro
  • 0.48.4 - Image support for Gemini 2.5 Pro
  • 0.48.5 - Change management for Cmd+Backspace -> Cmd+Shift+Backspace
  • 0.48.6 - Creates a new chat after the AI Pane has been closed for long enough
  • 0.48.7 - Fixes an issue where some chats could get deleted on update
  • 0.48.8 - Improves CPU performance from @-symbols search
  • 0.48.9 - Lower memory usage from indexing

0.47.x

Reliability, Keyboard Shortcuts & Early access opt-in

This release focuses mainly on stability and performance improvements to make existing features work better.

  • Memory Usage and Performance: The new release uses less memory and is more stable.
  • Keyboard Shortcuts: All keyboard shortcuts are now available in the Keyboard Shortcuts menu. Go to Settings > Keyboard Shortcuts to modify or add new shortcuts.
  • Early access opt-in: You can now opt in to early access from Settings > Beta > Update frequency.
  • Auto select model: We choose the most appropriate premium model for the task at hand based on performance, speed, and availability so you get performance even during model degradation or outages.
  • Themes: New and updated themes including Cursor Dark, Cursor Midnight, and Cursor Dark (High contrast).
  • UI improvements: Improved tool calling UI, thinking UI and error messages in chat. We've also added a new update notification in the app to make it clearer when a new version is available.
  • Rules: Allow nested .cursor/rules directories and improved UX to make it clearer when rules are being applied.
  • MCP: Added global server configuration with ~/.cursor/mcp.json and support for environment variables.
  • Sonnet 3.7 thinking: We've made a few improvements to 3.7 prompting. Thinking will now cost 2 requests instead of 1. More improvements will be rolling out in the coming days.
  • Ignore files: Improved .cursorignore behavior to more consistently exclude files
  • Upload multiple images: You can now upload multiple images at once in chat.
Patches (11)
  • 0.47.1: Improved performance, added back play button to apply code blocks
  • 0.47.2: Cursor Tab accepts work with single-line selections
  • 0.47.3: Fixes an issue with tool call errors on file edits
  • 0.47.4: Fixes an edge case where red diffs stick around in the editor
  • 0.47.5: Client-side support for upcoming MAX mode compatbility for Claude 3.7 Sonnet
  • 0.47.6: Faster applies, warns user when edits fails because of cursorignore
  • 0.47.7: MAX mode for 3.7 Sonnet with and without thinking
  • 0.47.8: UI fix for scrollable model name
  • 0.47.9: Performance improvements related to memory usage
  • 0.47.10: Fix for chat/apply not working with .github and other . prefixed folders, and small performance fixes.
  • 0.47.11: Release track picker (prerelease or standard), fix for dmg installer not working in rare cases

0.46.x

Agent is ready and UI refresh

  • Agent is ready: Agent is now the default mode, bringing a more powerful and unified AI experience. No more confusion between Chat, Composer, and Agent - just one smart interface that adapts to your needs.

  • UI refresh: Phase one of a fresh coat of paint with new default Cursor themes designed for focus. We've also simplified the @-context menu to make Agent more precise.

  • Web search: Agent can now automatically search the web for up-to-date information without requiring explicit @Web commands.

  • Ignore files: .cursorignore now blocks files from being added in chat or sent up for tab completions, in addition to ignoring them from indexing. We've introduced .cursorindexingignore for specifically controlling file indexing.

  • Agent tool limit: Upon reaching the 25 tool call limit, you can press "continue" to proceed (counts as a new request).

  • Project rules: - New capability to apply rules globally and a visual indicator for when rules will be applied

  • MCP improvements:

    • Agent can now automatically run MCP tools with Yolo mode
    • Configure Project Servers with <project-root>/.cursor/mcp.json
    • Support for Agent to use MCP resources as context
  • Tab rebinding: Tab suggestions can now be rebound to a different keybinding using editor.action.acceptCursorTabSuggestion

Fixes (3)
  • Crashes: Enhanced stability through memory and performance improvements, with additional crash prevention systems in place.
  • MCP: Improved reliability and quality of life enhancements for building MCP servers in Cursor.
  • Windows: Fixed newline errors in Chat Codeblocks and resolved auto-uninstallation issues.
Patches (11)
  • 0.46.1: Fixed HTTP2 and system certificate errors, resolved memory leaks
  • 0.46.2: Improved MCP reliability, added option to disable yolo mode for MCP
  • 0.46.3: Enhanced download reliability, fixed keybinding issue
  • 0.46.4: Added support for multiple image attachments and fixed "User aborted request" error
  • 0.46.5: Resolved issues with extended tool calls, long Composer sessions, and retry functionality
  • 0.46.6: Improved Windows installation process
  • 0.46.7: Improved memory usage and performance
  • 0.46.8: Fix for windows 'rename tools' update error, fix for some users unable to cancel agent terminal commands
  • 0.46.9: Memory and performances fixes, composer/chat remembers your last mode selection, http/1.1 support for agent/chat, linux python env vars fixed, cursorignore fixes
  • 0.46.10: Fix for chat/apply not working with .github and other . prefixed folders, and small performance fixes.
  • 0.46.11: Release track picker (prerelease or standard), fix for dmg installer not working in rare cases

0.45.x

.cursor/rules, Better Codebase Understanding, New Tab Model

  • .cursor/rules: Users can write several repository-level rules to disk in the .cursor/rules directory. The Agent will automatically choose which rule to follow.
  • Deepseek models: Deepseek R1 and Deepseek v3 are supported in 0.45 and 0.44. You can enable them in Settings > Models. We self-host these models in the US.
  • Summarize Previous Composers: When conversations grow too long, you can start a new conversation while referencing the previous one.
  • Agent sees recent changes: The agent can use a tool to see your recent changes. It also sees changes made between user messages.
  • Better Codebase Understanding: We've trained a new model for Codebase Understanding. We'll be rolling it out to all users on 0.45 in the coming week.
  • Fusion Model: We've trained a new Tab Model that is substantially better at jumps and long context. We'll also be rolling this out to users shortly.
  • Optional Long Context: When tagging long files, users have the option to request a larger context window with premium models. This will use more fast requests.

UPDATE (0.45.1-0.45.11): Fixes issue with older agent conversations, indexing stability, downloading incorrect extension versions, missing package on windows, crash on opening long composer sessions, latency on pasting code. Also adds MCP support, team-configurable blocklists, fixes Composer stuck on generating issue for some windows builds, exit code 5 crashes, and improves MCP UI. Also several improvements to memory usage.

UPDATE (0.45.12-13): Update infrastructure improvements. The F1 > "Check for Updates" command should now work.

0.44.x

Agent Improvements, Yolo Mode, Cursor Tab Update

  • Agent sees terminal exit codes, can run commands in the background, and commands are now editable
  • Agent reads linter errors to automatically fix issues
  • With Yolo Mode, the agent can auto-run terminal commands
  • @docs, @git, @web, and @folder are now available in the agent
  • Agent auto-saves changes to disk
  • Agent can decide to edit multiple locations in parallel
  • Agent can reapply edits with a smarter apply model
  • Composer changes and checkpoints are now persisted across reloads
  • Cursor Tab can make larger edits at a time
  • Better UX to review changes in Composer
  • 4o Support for Agent
  • Cheaper and faster Bug Finding Model

Bug Fixes

  • Fixed edge case where Cursor Tab crashed
  • Fixed stuck on generating bug in chat/composer
  • Composer no longer looks at files that were deleted
  • Fixed code selection edge-case for chat/apply
  • Cursor starts up faster

UPDATE (0.44.1-0.44.11): Fixes and improvements to dev containers, chat codeblocks on windows, and the agent. Decreases Cursor Tab Latency on Remote SSH. Fixes bug that prematurely triggered the free trial ended popup. Better observability for errors and crashes.