Autofix helps developers troubleshoot and resolve issues faster with AI-powered suggestions. It is currently available in Beta for Sentry users on a paid plan or trial.
Over the last two weeks, we have rolled out a number of new features and enhancements to Autofix. The most notable changes include:
- Autofix now leverages traces in its context: thanks to the inclusion of tracing data (spans and performance) in the context, Autofix has better outcomes when troubleshooting issues that impact distributed systems, spanning across service boundaries.
- More programming languages supported: with newly-added support for Bash, Dart, F#, PowerShell, Svelte, Swift, Vue, and Zig (among many others), you can now use Autofix on code written in almost every major–and not-so-major–programming language.
- Root cause analysis is more thorough: Autofix now performs a more comprehensive root cause analysis step, helping itself and human engineers identify problems.
- Additional project configuration: you can now indicate what repositories you would like to pass in as context, and customize repo-specific instructions for Autofix. These settings are accessible via the new gear button at the top of the Autofix flyout.
- More feedback buttons: you can now react with thumbs up/down on both the root cause and solution steps. This is incredibly helpful to our team as we evaluate Autofix’s performance and ship small improvements every day based on your feedback.
As we’ve announced earlier this month, we are also rolling out support for running Autofix automatically on issues in limited preview. With this new capability, Sentry will trigger Autofix on each issue that appears to be fixable automatically. Please send us a note at autofix@sentry.io to request access.
Among the most relevant bugs we’ve fixed:
- We've improved the speed and quality of codebase search and code editor.
- Q&A prompts from Autofix should now be more targeted and specific.
- The classification model powering automation decisions has been updated: Autofix will be more careful and targeted when deciding whether to automatically run on a new issue.
Thanks to all the users that have provided feedback on Discord!
Github
Slack
Javascript
Python
React
Laravel
Next.js
