Opslane combines browser error reporting with session recordings so an AI worker can investigate problems users actually encounter. The project also targets failures that produce no exception, such as a button that does nothing or a form users repeatedly abandon.
It groups related issues and ranks them by affected users before deciding which to investigate. The worker then reads the repository and proposes a fix. Its documented verification step reruns the project’s build and tests, followed by a second model’s review, before opening a pull request. Unverified fixes return a reason for human attention instead.
The self-hosted stack keeps state in Postgres and recordings in compatible object storage. Investigations still use outside services: Anthropic for the model, E2B for test sandboxes and GitHub for code access and pull requests. Browser inputs are masked by default.
The full workflow currently targets JavaScript applications. Opslane is AGPL-licensed, while its browser and Python SDKs use MIT.