Bubo stays quiet until it has evidence, and speaks only when something's worth your time. Your LLM, self-hosted. No filler. No praise. No generic summaries.
Capabilities
Bring-your-own-LLMSelf-hostedGitLab + GitHubSignal over noiseReview moodsDispute-driven learningVerify before postingGovernance & provenanceOpenTelemetryMCPGitHub Action CI
Features
Reviews worth reading
Only what mattersWhat’s wrong, why, and how to fix it. No praise, no summaries.
Dry runRuns a real review and enumerates findings, but posts nothing.
Pick the toneTerse, collaborative, socratic, formal, or casual.
Use the AI you already trust
Bring your own modelCodex, Claude, or any CLI-driven model. No lock-in.
SCMWorks with GitLab and GitHub.
Decide what gets flagged
Gate or collaborateMerge-blockers only, or everything — suggestions and questions too.
Confidence calibrationPost only what it’s sure of; raise the bar per issue type.
Learns your tasteKeeps flagging what you dismiss? It stops.
Double-checks firstAn optional second model drops findings that don’t hold up.
Fits how you work
MCPConnect your editor to trigger an interactive review on demand.
Reviews in CIA GitHub Action that comments on PRs in your pipeline.
A simple dashboardRead-only view of recent reviews, health, and reports.
Governance, compliance & ROI
AI-code provenanceIdentifies AI-assisted and sensitive changes for closer review; it does not block merges by itself.
Audit trailA write-once, chronological record of every decision and its lifecycle.
Self-hosted controlRepository data stays within the infrastructure and model path you configure.
ROI reportingAccept/dispute rate, bugs caught, noise, and cost.
Redaction and signed releasesCredentials scrubbed before the model sees the diff. Signed releases, SBOM.