Launching today
heal.dev
Find bugs and vulnerabilities code review misses
140 followers
Find bugs and vulnerabilities code review misses
140 followers
Code review catches what’s wrong in a diff. Heal catches what breaks when the app runs. Heal learns how to run your application, builds a production-like sandbox with external services stubbed, then hunts for bugs, performance regressions, and runtime vulnerabilities users would hit. It reproduces every defect before reporting it, so you get real bugs to fix, not a backlog of AI-generated suspicions.
Free Options
Launch Team / Built With
BoxLite Open SourceOpen-source microVMs for AI agents, embedded in your app
Promoted
Reproducing every defect before reporting it is the right bar, most review bots drown you in maybes. How do you handle apps where the bug only shows up with real third party responses, like a payment or email provider returning something odd that your stubs don't model?
@thamibenjelloun That's a great question. The agent adds the best stubs it can (it can take docs as an input), but if you need to add more knowledge about something that happened and isn't in the mocks, you can tell the agent (eg. with our MCP server) to add it.
We had this use case on a side project with Maptiler and Strava integrations, logged the errors in @BetterStack with their MCP server, and asked the agent to enrich the mocks with them.
@thamibenjelloun also the agent starts to ask you to review the local sandbox it'll build, such that you can explicitly tell it if you want to use a real third-party / provider with staging/dev access.
How much configuration does it take for Heal to learn how to run an app? For projects with non-trivial local setups (multiple services, env specific config), that first step seems like the real test.
@mehdi_hamri Hey! A couple of metrics on that: small codebase (one frontend app, a backend app and a single database) up to 20min. For a large codebase with multiple frontends, backends and a lot of storage and external dependencies, the agent takes the time it needs (we've seen 3 hours for an enormous codebase). But, the agent is train to stop and ask the user for more informations when it's blocked or judges it can speed up the build with the user insights.
Reproducing a defect before reporting it sets a useful bar. I would like to see the smallest failing case alongside each finding.
@oleksii_sekundant spot on! What heal does is give a script/test that allows to reproduce the issue and validate it's fixed. Both useful for repro, and having a feedback loop to verify the work of the coding agent you ask for a fix.