↓Skip to main content
  1. Tools/
  2. AI Sessions/

roborev

·2 mins·
ai-ready AI Sessions external Ai-Sessions Code-Artifacts Code-Review Git-Hooks Audit Sqlite Ai-Agents Accountability
Language: Go
License: MIT
Maturity: stable
Verified: 2026-08
Table of Contents

roborev is a continuous code review system that installs as a git hook and automatically reviews commits generated by AI coding agents. It maintains a local review database and is designed to introduce accountability into agentic coding workflows where large volumes of AI-generated commits would otherwise go unreviewed.

How It Works #

On each commit, roborev:

  1. Extracts the diff and commit message
  2. Submits them to a configured LLM for review
  3. Stores the review result in a local database (~/.roborev/)
  4. Optionally applies auto-fixes or feeds findings back to the agent

Reviews are timestamped, linked to commits, and queryable. The review database persists across sessions so the full history of agent activity and review outcomes is retained.

Storage #

  • Primary: local SQLite at ~/.roborev/
  • Optional: PostgreSQL for team/shared environments
  • Export: roborev export reviews produces JSON for downstream use

Telemetry #

roborev sends anonymous usage statistics to PostHog (review count, feature flags) but explicitly excludes repo names, paths, remotes, prompts, review output, provider tokens, usernames, and IP geolocation.

CI Integration #

roborev can export benchmark and CI metrics, and optionally integrates with kata for correlating review findings with issue tracking.

Relevance to Archival #

roborev creates a provenance and accountability trail for AI-generated code: which agent authored which commit, what issues were flagged in review, and whether they were addressed. This is complementary to session-level archival tools (ccexport, entire-io) that capture the conversation producing the code. Together they provide both the dialogue and the quality-assurance record.

The review database is a natural candidate for inclusion in a DataLad dataset alongside the code repository it describes, though no established workflow for this exists yet.

See Also #

  • kata – local issue tracking for AI-assisted work
  • agentsview – session analytics across 40+ agents
  • git-ai – line-level AI authorship attribution via git notes
  • entire-io – git-native session archival