Skip to content
Investigate

AI analysis & fixes

Use captured evidence to investigate a likely cause and review proposed changes.

What analysis uses

AI analysis can work from captured errors and request context without a connected repository. With source access, it can inspect relevant code and propose changes. Link the correct repository and full deployed commit so the analysis does not inspect an unrelated revision.

Run and review an analysis

  1. Open the error or incident you are investigating and choose its analysis action.

  2. Check that the relevant occurrence, node, and source revision are available.

  3. Read the proposed cause alongside the request evidence and cited source locations.

  4. Review suggested changes as you would any other patch. Check that they address the observed failure and preserve expected behavior.

Validation and pull requests

Cloud workflows can validate a suggested patch in an isolated environment and open a pull request when the workspace's PR policy permits it. Inspect which checks ran and whether they passed, failed, or were unavailable.

Enterprise Agent analysis stores its reports and proposed patches locally. Automatic PR publishing and application of a patch are separate workflows; do not assume an agent analysis publishes or deploys a fix.

Usage and data handling

Cloud analysis uses AI credits. Check your workspace's Billing page for the allowance and extra usage costs. For self-hosted execution with your own model credentials, the configured provider bills model usage and receives the inputs needed for analysis.