2025

• Historical cycles chart now displays 12 months of data (up from 5 months)
• New export functionality for previous billing cycles, enabling better cost analysis and reporting

• Support for modifying default commands like /plan and /review
• Ability to create entirely new custom commands tailored to your team’s workflows
• Command management interface for enterprise administrators

• In-thread responses with updates and questions to help with software engineering tasks
• Support for mapping Teams channels to specific Devin organizations

• The “Connected accounts” section in organization settings has been renamed to “Integrations”.

• Duplicate detection: Requests containing identical permissions (same connection, repository path, and group prefix) are now rejected with a clear error message listing the duplicates.
• Overlap detection: Requests containing a repository-specific permission that conflicts with a group prefix permission are now rejected. For example, adding both group_prefix: "myorg" and repo_path: "myorg/repo1" for the same connection will fail, since the group prefix already covers that repository.

• The maximum pagination limit for Enterprise API v2 query parameters has been reduced from 1000 to 200 for improved performance and reliability.
• This change affects all v2 Enterprise API endpoints with pagination, including sessions, members, organizations, groups, and user usage endpoints.
• The default limit remains 100, with a minimum of 1 and a new maximum of 200.
• Note: This change does NOT affect the v1 External API.

• Product polish and design system migration for playbooks and settings pages
• Repository page improvements with better filtering and pagination
• Ongoing bug bashing and quality of life improvements

• Fixed bug preventing proper display of secrets page for organization members
• Fixed bug around gh CLI auth failing in certain circumstances on Devin’s machine
• Enterprise API v2 git permissions endpoint now returns permission_names dict mapping permission IDs to repo/group paths
• Enhanced member page tabs and navigation
• Ongoing stability improvements and bug fixes

• Fixed positioning of “Learn more” tooltip and link in message components by moving it inside a div container for better layout structure
• Removed login_hint in new user authentication flow due to inconsistent behavior, preventing email from pre-filling during a user’s first visit.
• Fixed a bug where you could not add more than one set of session-scoped secrets.

• We’ve launched the DeepWiki MCP server, providing programmatic access to DeepWiki’s repository documentation and search capabilities.
• Connect your AI applications to DeepWiki using the Model Context Protocol (MCP) standard.
• Free, no authentication required.

• At multiple points in each session, Devin will express its confidence:
  • At the start of the session
  • After creating a plan
  • Whenever it answers a question about the code
• When Devin doesn’t have 🟢 confidence (i.e., 🟡 or 🔴), it will now wait for user approval before proceeding with its plan. If it’s 🟢, it proceeds automatically.
• You can still send feedback and adjust Devin’s plan even after it has started working.
• Our data shows that Confidence Scores are highly correlated with success.

• Easily get Confidence Scores for multiple issues at once directly through our Linear integration and Jira integration.
• This occurs without starting actual Devin sessions, allowing you to score as many issues as you’d like and prioritize having Devin work on the highest-confidence tasks.

• The codebase understanding and intelligence features from DeepWiki are now directly integrated into Devin.
• At any point during a session, ask a question, and Devin will provide a DeepWiki-like answer, complete with code citations.
• Devin auto-detects when it should scan your codebase, but you can also trigger this manually using !ask.

• Add the “Devin” label to a Jira issue, or bulk add to multiple issues to begin issue scoping in parallel.
• Devin knows your codebase and will comment on each ticket in minutes with a summary of relevant code, implementation plan and open questions.
• Use Devin’s analysis to get up to speed. Devin has the self-awareness to report 🔴/🟠/🟢 confidence estimates too.

• Devin will now link to Devin Spaces in its comment on Jira and Linear - use Spaces to ask follow up questions and further plan out the task.
• Start a Devin session directly from the plan you create in Devin Spaces!

• Use Cmd+A to multi-select Linear tickets. Then, add the Devin label to start ticket scoping in parallel.
• Devin knows your codebase and will comment on each ticket in minutes with:
  • A summary of the current code
  • An implementation plan
  • Any edge cases or questions that need your attention
• Use Devin’s analysis to get up to speed, or just click the provided link to have Devin take a first pass at the PR. Devin has the self-awareness to report 🔴/🟠/🟢 confidence estimates too.

​

Devin IDE

• Click “Review Changes” for a diff view of file edits so far. You’re in a fully featured IDE, so you can open files in new tabs, jump to definition, etc.

• Devin may send citations or references to code. Clicking on these will deep link into VSCode!
• Click “Follow Devin” to follow Devin’s edits in real time. Click stop to take over and use the IDE yourself.
• Use ⌘K to generate terminal commands from natural language.

• Use ⌘I for rapid responses to questions or rapid file edits.

• All of Devin’s terminals, commands, and their outputs are available in VSCode. Toggle from read-only to writable to run your own commands.

• Test and fix changes end to end without leaving the Devin webapp. Ask Devin to run your app locally, or take over and run commands yourself. Then use Devin’s browser to test the local build yourself!

​

Interactive Planner

• Devin rapidly scans relevant files and code snippets to generate an initial plan. This initial plan and subsequent messages may now cite code snippets and files, and clicking on these citations now deep links into the Devin IDE!
• For more complex tasks, click “Wait for my approval” so that Devin waits for your feedback on its full plan. Brainstorm and explore the codebase together in VSCode to refine the plan.
  • By default, if you don’t click “Wait for my approval”, Devin waits 30 seconds for your input before proceeding. You can always change the default behavior in Settings > Customization.

​

Ask Devin

• 🔎 Ask Devin -> Devin: Ask Devin to make code changes after using Ask Devin to find the relevant code. Use Cmd + Enter to quickly construct a high quality Devin prompt using your search context.
• 🔬 Deep Mode: Toggle on deep mode for complex questions that require extended research.
• 📓 DeepWiki is used by Ask Devin to better understand your codebase, and may help you as well! It contains architecture diagrams, links to sources, and more. Check it out at the bottom of your sidebar.
• 💬 Ask follow up questions - you can scroll up/down or use the component on the right (appears on hover) to navigate your history
• 🔗 Share your Ask Devin results: Try sharing a link to your search results when discussing code with your co-workers
• 💡 Tip: For now, we recommend setting up a Site Search Shortcut in Chrome so you can more quickly start Ask Devin queries from your browser address bar. Just go to chrome://settings/searchEngines and add a site search with url http://app.devin.ai/search?prompt=%s

• Connect multiple GitHub orgs to the same Devin account - you can easily set this up in Settings > Organization Integrations. Let us know if you’d like this enabled for your team.
• Manage all of Devin’s tasks in the new Session Manager. Easily filter for sessions by PR status, users, playbooks and export session data.
• Tag sessions with custom tags that you can filter by in the Session Manager. Click on the 3 dots icon of the session page to “Edit Tags”.
• Customize when Devin auto-closes PRs due to inactivity under Settings > Customizations.
• Easily tag @file_name to reference files in Devin’s input box so Devin can quickly find the right place in your codebase to review and/or edit. Note that this only works for files in repos that have been set up in Devin’s Workspace.

• Speed: Devin is ~2x faster vs in October 2024 and takes ~7.8 minutes on average to complete junior developer tasks in our internal evaluations.
• Copy/paste in Devin’s browser: You can now copy text from your browser and paste it into Devin’s browser! This was a highly requested feature that removes a major friction with providing Devin access to accounts (you no longer need to type in your passwords)!
• Helping users with their prompts: Devin proactively gives you feedback on suboptimal prompts & proposes breaking tasks down when they’re too complex.
• Gitlab (beta): Connect both Gitlab and GitHub repos to Devin! Devin can now push, pull, and view/create Gitlab MRs. Contact us via app.devin.ai/settings/support to set this up.
• Batch edits: Tell Devin to “find and edit” code to encourage Devin to “fan out” and edit an arbitrary number of files in parallel. This greatly improves speed, especially for repetitive refactors.

• Multi-action: Devin can choose to perform any set of diverse batch of actions optimistically (e.g. viewing the browser, while running a shell command, while reading 10 code files), improving speed.
• Browser improvements: We’ve shipped browser changes that allow Devin to:
  • deal with auto-opening tabs (required for some complex auth flows)
  • use multiple tabs (helpful for iteratively comparing 2+ webpages)
• Local UI Testing: Devin can better test + visually understand UI changes locally.
• Customize chat vs workspace width: Drag to make the chat as narrow or wide as you’d like! The editor in the workspace is also easier to navigate now, with file tree on the left.
• Repo setup (in Devin’s Workspace): We verify that all the commands you provide Devin (to run lint, install dependencies, and run tests) run successfully, and Devin will surface in chat if any of these commands don’t succeed.
• Sonnet 3.7 in Devin: We incorporated Sonnet 3.7 in Devin on 2/24, with optimizations to our use rolling out starting 2/26. In our testing, the new model is the best we have seen to-date on a variety of tasks including debugging, codebase search, and agentic planning.
• Keyboard shortcuts: Use → ← or ↑ ↓ anywhere on the session page to step through Devin’s workspace progress over time.
• Devin PR Metrics: app.devin.ai/metrics now shows all PRs opened by Devin, even when 2+ PRs were opened in the same session.
• Faster startup: Devin only installs dependencies for the repositories needed in a session.
• Addressing your PR review feedback: Devin is more reliable at remembering to address all the review comments you left on its PR.
• Misc brain improvements: Devin is less likely to loop while trying to fix CI/lint failures, is better at planning, is better with git, and many more improvements!

• each action Devin took (e.g. “Edited github.py”)
• Devin’s thoughts explaining why the action was taken (e.g. “There was a type error….the fix involves XYZ”)
• any editor diagnostics errors present after the action was taken (in red)

• Manage members and access controls for all organizations
• Centrally manage billing across all organizations

• Navigate to Settings > Devin’s Workspace > Danger Zone
• Switch to Large Performant (Beta) - this will require resetting your machine setup. If you want to opt in to experimental auto-migration, reach out to support@cognition.ai or via Slack Connect

• Keeping sessions under 10 ACUs (Devin’s performance degrades in long sessions)
• Providing details in your very first instruction to Devin, including (1) specific requirements (2) high level description of the task (3) what Devin should do after making the requested changes - e.g. testing instructions, PR guidelines, or tell Devin to wait for CI to pass without testing locally
• If you find yourself often re-using instructions, add them to Devin’s knowledge in Settings > Devin’s Settings > Knowledge