Bubo

Bubo reviews GitHub pull requests and GitLab merge requests using an AI model you choose. It posts findings beside the relevant code, with an explanation of the problem and a suggested fix. You can run a review without posting anything, adjust confidence thresholds and limit comments to defects rather than general suggestions.

The review service polls your repositories and keeps review state in SQLite. A dashboard shows recent reviews and health, while reports track accepted or disputed findings, latency and estimated model costs. An MCP server lets you trigger reviews and inspect results from compatible tools.

You can install Bubo as a Python package or use its Docker image. The current package requires Python 3.14 or newer, and the review agent CLI needs separate installation and authentication. That setup suits teams comfortable managing repository tokens, model access and a service on their own infrastructure.

The software is free under the MIT licence. Hosting and AI usage remain your responsibility, and cost reports depend on the rates you configure. Hosting Bubo yourself does not keep code local if your chosen model sends it to an external provider. Anonymous usage analytics are enabled by default and can be disabled. Start with posting turned off, inspect the findings and keep human review in the process.