AI pull-request reviewer that runs locally or in CI pipelines
Run automated, model-driven pull request reviews that analyze git diffs and flag logic, security, and bug patterns before merge. Matt Coles, the developer, targets code reviewers and DevOps teams in need of low-friction automation. The tool supports multiple cloud LLM providers and local models via Ollama, offers keyless OIDC authentication, and integrates with GitHub Actions, GitLab CI, or a local CLI workflow for pre-push checks.
What does lgtmaybe actually check in a pull request?
lgtmaybe inspects git diffs and generates structured findings, using an automated diff compression step to remove generated files and binaries so the analysis focuses on logic changes. It validates output against a ReviewFinding JSON schema, which yields machine-readable results you can consume in CI or display in PR comments. The pipeline supports both cloud and local LLMs, so reviews can run against different model providers with a single flag.
How does it affect system and CI performance during reviews?
The tool can run as a local CLI or inside CI, so system impact depends on the chosen mode. When used locally with Ollama or OpenAI-compatible endpoints, the review runs on the developer machine; in CI it calls provider APIs. The automated diff compression reduces payload size before model calls, which lowers network and processing overhead in hosted pipelines compared with sending full repositories.
Is lgtmaybe safe to use in corporate or cloud environments?
The developer built a keyless authentication option using OIDC for cloud providers, avoiding static API secrets in CI environments. In addition, the design permits running reviews entirely locally with open-weight models, which prevents code from leaving the host. The tool's gate step filters out low-confidence findings to reduce noisy or speculative suggestions from appearing in automated reports.
Does it require deep technical knowledge to operate correctly?
The CLI exposes modes for local diffs and CI integrations, so basic use involves running the pipeline over a git diff or enabling the provided GitHub Action. Advanced setups that switch providers or run local models require configuring Ollama or provider endpoints. The JSON schema output aids automation, but parsing or integrating those results into custom tooling needs developer familiarity with CI scripting.
A practical reviewer for teams that need flexible, secure AI feedback
lgtmaybe is a suitable option for developers and DevOps engineers who want AI-assisted PR analysis without obligating cloud secrets, because it supports keyless OIDC and local model runs. Its main limitation is the operational overhead for teams that must configure local model infrastructure or parse the JSON findings into custom workflows, which adds setup work before it fits smoothly into existing pipelines.
