Baz Reviewers

Code review agents that analyze pull requests for quality, correctness and security issues.

Baz default reviewers are AI agents that evaluate pull requests in repository and external context rather than only per-file. Reviewers split the codebase into indexable units, use embeddings and similarity to retrieve relevant code and tests, and run agents that combine static analysis and optional runtime inspection to produce structured findings. Findings are surfaced as pull request comments and as CI check results.

Reviews work by retrieving relevant code, tests and fixtures via embeddings and similarity, running an agent with tools for ticket and design access, image handling and browser automation, and persisting results as queued jobs that carry pull request metadata.

Code review agents maintained by Baz

Agent spotlight: Logical Bugs

The purpose of this reviewer is to detect complex, cross-file logical errors that static linters miss.

• Technical attributes:

  • Uses whole-repo retrieval: pulls call graphs, helper functions, and test fixtures via similarity search to reconstruct the logical context for a change.

  • Combines AST-based program understanding with agentic reasoning to propose concrete execution scenarios that violate invariants (for example, inconsistent state transitions, missed early returns that lead to surprising side effects, or incorrect assumptions about data shapes).

  • Correlates dynamic evidence (if available) such as existing test traces or recorded failing CI logs to strengthen hypotheses.

  • Outputs Findings that include the reasoning trace, implicated files and lines, and suggested repro steps or minimal code locations to inspect.

  • Constraints: this agent is compute-heavy and tuned to reduce false positives by requiring multi-source evidence (retrieval + AST +, where possible, runtime traces).

Agent spotlight: Spec Reviewer (Beta)

The purpose of this reviewer is to validate that a change meets ticket and design requirements by combining PR diff, Jira ticket context, Figma designs, and a live preview environment.

• Technical attributes:

  • The agent is automatically initated when a pull request is linked to a Jira ticket and optionally a live preview environment. If the ticket contains references to a Figma design the agent will include it in its context.

  • The agent builds a sandboxed browser session and delegates UI interactions and visual checks to sub-agents. The agent runs test cases, compares rendered UI to design artifacts, and verifies ticket requirements.

  • Output: a single PR comment grouping unmet requirements first and met requirements collapsed, as well as a GitHub check with pass/fail/neutral.

  • Constraints: requires a reachable or credentialed preview environment. Inline annotated screenshots are planned but not yet implemented. The agent is designed to reduce noise by summarizing issues in one place rather than producing many inline comments.

FAQ

What do Baz’s default (managed) reviewers do?

They analyze change requests for naming, typing, logic bugs, outdated comments, log errors, etc., using a combination of AI, parsing, and repository context.

How does Baz scale efficiently on large codebases?

Baz divides code into manageable chunks, reprocesses only changed files, stores embeddings in a vector database for similarity search, and filters by organization/repo to maintain performance.

Can I disable some default reviewer checks?

Yes. Organization admins can deactivate specific agents or modify their scope.