Cursor agents
ComponentUpdates related to the agents component of Cursor.
All AGENTS Features
- Introduced a Tiled layout in the Agents Window, allowing users to split the view into panes to run and manage several agents in parallel.(3.1)
- Enabled users to press and hold Ctrl+M to activate upgraded, more reliable voice input with batch STT transcription.(3.1)
- Added UI controls (waveform, timer, cancel/confirm buttons) for voice recording sessions.(3.1)
- Allowed searching and selecting a branch before launching a cloud agent from the empty state, targeting the agent to a specific branch immediately.(3.1)
- Enabled direct navigation from a diff view straight to the exact corresponding line within the file.(3.1)
- Added include/exclude filters to the "Search in Files" functionality for scoping code searches to specific file sets.(3.1)
- Introduced the new Agents Window interface allowing users to run many agents in parallel across repos and environments (locally, worktrees, cloud, SSH).(3.0)
- Enabled Design Mode within the Agents Window to annotate and target UI elements directly in the browser for more precise feedback.(3.0)
- Added keyboard shortcuts for Design Mode: → ⌘+Shift+D to toggle, Shift+drag to select an area, → ⌘+L to add element to chat, and → Option+click to add element to input.(3.0)
- Added Agent Tabs in the Editor to view multiple chats side-by-side or in a grid.(3.0)
- Added a new command /worktree that creates a separate git worktree for isolated changes.(3.0)
- Added a new command /best-of-n that runs the same task in parallel across multiple models in isolated worktrees and compares outcomes.(3.0)
- Enabled MCP Apps to support structured content for richer tool outputs.(3.0)
- Added directory group name to audit logs for improved human readability.(3.0)
- Added a team-level Admin setting to restrict creating, editing, and deleting team secrets to Admins.(3.0)
- Added an Enterprise Admin control for disabling "Made with Cursor" code attribution for the entire team.(3.0)
- Introduced support for self-hosted cloud agents, allowing code execution and tool usage entirely within your private network.(03-25-26)
- Enabled self-hosted cloud agents to be initiated from the Cursor editor, web app, Slack, GitHub, Linear, or via the API.(03-25-26)
- Enabled Cloud Agents to use the software they create for testing changes and demonstrating work.(02-24-26)
- Introduced isolated Virtual Machine (VM) environments for each Cloud Agent running on your codebase, complete with a full development environment.(02-24-26)
- Cloud Agents now produce merge-ready Pull Requests (PRs) that include artifacts like videos, screenshots, and logs for easier review.(02-24-26)
- Introduced the Cursor Marketplace for discovering and installing plugins to extend Cursor's capabilities.(2.5)
- Enabled granular network access controls for sandboxed commands, allowing users to define allowed domains (User config only, User config with defaults, Allow all).(2.5)
- Enabled asynchronous execution for subagents, allowing the parent agent to continue working while subagents run in the background.(2.5)
- Enabled subagents to spawn their own subagents, supporting complex tasks like multi-file features and large refactors.(2.5)
- Introduced long-running agents capable of planning and completing larger, more complex tasks autonomously.(02-12-26)
- Enabled long-running agents to deliver larger, more complete Pull Requests (PRs) with fewer follow-up tasks.(02-12-26)
- Introduced Subagents, specialized, independent agents that run in parallel with custom configurations for faster, more focused task execution.(2.4)
- Enabled direct image generation from the agent using text descriptions or reference uploads, saving previews inline and to the assets/ folder.(2.4)
- Launched Cursor Blame (Enterprise plan feature) extending git blame with AI attribution, showing which lines were AI-generated (by Tab completions or agent runs) versus human-written.(2.4)
- Added the ability for agents to ask clarifying questions during Plan and Debug modes, allowing them to continue working while waiting for user input.(2.4)
- Enabled building custom subagents and skills that utilize the new 'ask question tool'.(2.4)
- Added the agent models command to list all available models.(CLI)
- Enabled quick switching between models using the --list-models flag or the /models slash command.(CLI)
- Introduced the /rules command to create new rules and edit existing ones directly from the CLI.(CLI)
- Enabled enabling and disabling MCP servers on the fly using the /mcp enable and /mcp disable commands.(CLI)
- Introduced four default layouts: agent, editor, zen, and browser for easier customization.(2.3)
- Enabled switching between layouts using the keyboard shortcut Command (⌘) + Option (⌥) + Tab (⇥).(2.3)
- Enabled moving backwards through the layout list by including Shift (⇧).(2.3)
- Introduced Conversation Insights to analyze agent sessions, categorizing work type (Bug fixes, refactoring, explanation) and complexity.(enterprise-dec-2025)
- Enabled sharing of agent conversations via read-only transcripts, which can be forked to start new conversations from the same context.(enterprise-dec-2025)
- Added Billing Groups functionality for enterprise users to map usage and spend to organizational structure, set budget alerts, and track team adoption.(enterprise-dec-2025)
- Extended Linux sandboxing support for agents, complementing existing macOS support, allowing scoped access control.(enterprise-dec-2025)
- Launched Service Accounts for non-human automation, enabling secure configuration, API calls, and cloud agent invocation without tying workflows to individual developer accounts.(enterprise-dec-2025)
- Introduced Debug Mode which instruments the app with runtime logs to help reproduce and fix tricky bugs across stacks, languages, and models.(2.2)
- Enabled a new browser sidebar and component tree for simultaneous design and coding, allowing real-time element manipulation, color updates, layout testing, and CSS experimentation.(2.2)
- Added the ability to select multiple visual elements and describe desired changes in text to initiate an agent for visual updates.(2.2)
- Added support for inline Mermaid diagrams in Plan Mode, allowing agents to automatically generate and stream visuals into plans.(2.2)
- Enabled sending selected to-dos to new agents for more granular control over plan building.(2.2)
- Implemented Multi-agent judging, which automatically evaluates parallel agent runs and recommends the best solution with an explanation.(2.2)
- Introduced AI Code Reviews directly in the editor to find and fix bugs within your changes via a sidepanel.(2.1)
- Enabled Instant Grep (Beta) so that all agent-run grep commands, including manual sidebar searches and regex searches, execute instantly across all models.(2.1)
- Introduced a new Multi-Agents editor featuring a sidebar for managing agents and plans.(2.0)
- Enabled running up to eight agents in parallel on a single prompt using isolated environments (git worktrees or remote machines).(2.0)
- Introduced Composer, a new agentic coding model that is 4x faster than similarly intelligent models.(2.0)
- Launched Browser functionality to General Availability (GA) with added support for Enterprise teams.(2.0)
- Enabled Browser to be embedded in-editor, including new tools for element selection and forwarding DOM information to the agent.(2.0)
- Enabled defining custom commands and rules for your Team via the Cursor dashboard, automatically applied to all members.(2.0)
- Introduced Voice Mode, allowing control of the Agent using built-in speech-to-text conversion.(2.0)
- Enabled defining custom submit keywords in settings to trigger the agent to begin running in Voice Mode.(2.0)
- Enabled creating plans with one model and building the plan with another, with options for foreground, background, or parallel agent execution.(2.0)
- Enabled sharing custom rules, commands, and prompts with the entire team via deeplinks.(2.0)
- Enabled Enterprise teams to distribute hooks directly from the web dashboard, including managing drafts and OS targeting.(2.0)
- Enabled viewing a timestamped Audit Log of admin events including user access, setting changes, and team rule edits.(2.0)
All AGENTS Bug Fixes
- Fixed stuttering when scrolling through long agent conversations.(3.1)
- Fixed an issue where an agent conversation containing many diffs or code blocks would flash and freeze.(3.1)
- Fixed text area behavior for Network Access Controls so pressing Enter reliably adds a newline at the end of the input.(3.0)
- Fixed hooks loading so multi-root workspaces correctly read project hook files from all workspace folders instead of only the first one.(3.0)
- Fixed a markdown parsing bug where parenthesized HTTP(S) links could be misread as citations.(3.0)
- Fixed todo card visibility to prevent them from disappearing after all todos complete.(3.0)
- Fixed Composer queued prompts that were not resuming automatically after editing operations.(3.0)
- Fixed picker behavior for disabled but selectable models by removing misleading styling and auto-enabling the model upon user selection.(3.0)
- Fixed a bug where expanding/collapsing thinking blocks didn't work while streaming was still in progress.(3.0)
- Fixed a bug where Shift+Enter line breaks weren't treated as multiline content, preventing the prompt input field from staying in an incorrect state.(3.0)
- Fixed issues related to synchronous execution blocking the parent agent when using subagents.(2.5)
- Fixed 11 unspecified bugs (as noted in the release notes summary).(2.4)
- Fixed unspecified bug fixes related to CLI operations.(CLI)
- Fixed file conflicts when running multiple agents in parallel by using isolated environments.(2.0)
- Fixed memory leaks and improved overall memory usage across the application.(2.0)
- Improved quality when working with Agent across all models, especially GPT-5 Codex, due to harness improvements.(2.0)
Releases with AGENTS Changes
3.16 features2 fixesThis release significantly enhances the Agents Window with a new Tiled layout for multi-tasking and introduces upgraded, more accurate voice input with dedicated UI controls. Users can now also select branches directly when launching cloud agents and benefit from improved performance and stability across chat scrolling, file tree updates, and large edit streaming.
3.010 features8 fixesCursor 3 introduces the powerful new Agents Window, enabling users to run multiple agents in parallel across various environments with improved visibility and control. This release also adds Design Mode for precise UI targeting and Agent Tabs for managing multiple chats simultaneously. Several usability improvements focus on performance, such as faster large-file diff rendering, alongside key bug fixes across input handling and workspace loading.
03-25-262 featuresThis release introduces self-hosted cloud agents, a major feature allowing users to run code execution and tool calls entirely within their own infrastructure for enhanced security and privacy. These agents offer the full feature set of hosted agents and can be managed across multiple platforms including the editor, web app, and integrations.
02-24-263 featuresThis release introduces powerful Cloud Agents that operate within isolated VMs to build, test, and demo software directly from your codebase. Agents now deliver merge-ready Pull Requests complete with review artifacts, and are accessible across all Cursor platforms.
2.54 features1 fixThis release significantly enhances Cursor's extensibility with the introduction of the Cursor Marketplace for plugins, covering workflows from design to deployment. Key functional updates include enabling asynchronous subagents for better parallel task handling and introducing granular network access controls for sandboxed commands for improved security.
02-12-262 featuresThis release introduces long-running agents that can autonomously complete larger, more complex tasks by planning ahead. These new agents are now available for Ultra, Teams, and Enterprise users and result in more complete PRs.
2.45 features1 fixThis release significantly enhances agent capabilities by introducing Subagents for parallel, specialized task execution and enabling direct image generation within the agent interface. Additionally, Enterprise users gain Cursor Blame for detailed AI attribution tracking, and agents can now proactively ask clarifying questions during complex workflows.
CLI4 features1 fixThis release significantly enhances the Command Line Interface (CLI) by introducing new controls for managing models, rules, and MCP servers directly via new slash commands and flags. Users can now list models, manage rules, and toggle MCP servers on the fly, alongside notable performance improvements for CLI hooks.
2.33 featuresThis release focuses on stability improvements across the core agent, layout controls, and code review functionality. Users can now easily customize their workspace layout by selecting from four new default layouts and switching between them using a new keyboard shortcut.
enterprise-dec-20255 featuresThis release focuses heavily on Enterprise features, introducing Conversation Insights to analyze agent work types and the ability to share agent transcripts with teams. New administrative controls include Billing Groups for usage tracking and Service Accounts for secure automation. Additionally, agent sandboxing now supports Linux environments.
2.26 featuresThis release introduces powerful new debugging capabilities with Debug Mode and a visual development workflow via the new browser sidebar and component tree. Plan Mode is enhanced with inline Mermaid diagram support, and Multi-agent judging provides automated selection of the best solution when running agents in parallel.
2.12 featuresThis release introduces powerful new capabilities including AI Code Reviews directly in the editor for immediate bug detection and Instant Grep (Beta) for lightning-fast codebase searching. Plan Mode has also been enhanced with interactive question-answering to improve plan quality.
2.012 features3 fixesThis release introduces significant new capabilities, including the Multi-Agents editor for running up to eight agents in parallel and the new, faster Composer coding model. Key stability improvements include Sandboxed Terminals going GA on macOS and major performance boosts for Language Server Protocols. Users can now also control the agent via Voice Mode and define centralized Team Commands.