Skip to content

Pull request review you can read the source of.

open-nitpick reviews pull requests with the model provider you choose and can post inline comments. Run it as a GitHub Action, in any CI system, from an agent, or against local changes before a pull request exists. The reviewer is self-hosted; review input is sent to the model provider you configure. Choose a local provider to keep review input on your machine.

Get started CLI reference Set up CI

Review changes · Configure models and policy · Use from an agent · All workflows · Trust model

Read the provider guidance before selecting a hosted endpoint, inspect the measurements, or read the source on GitHub.

# A release binary, or `go install …/cmd/nitpick@latest` with Go 1.26.0.
v=$(gh release view --repo jdziat/open-nitpick --json tagName -q .tagName)
os=$(uname -s | tr 'A-Z' 'a-z'); arch=$(uname -m | sed 's/x86_64/amd64/;s/aarch64/arm64/')
curl -fsSLo nitpick "https://github.com/jdziat/open-nitpick/releases/download/$v/nitpick_${v}_${os}_${arch}"
chmod +x nitpick && sudo mv nitpick /usr/local/bin/

export LLM_PROVIDER=openrouter
export LLM_MODEL=z-ai/glm-5.3-flash
export OPENROUTER_API_KEY=sk-or-...

nitpick review          # reviews your uncommitted changes

Bring your own models

18 providers, any OpenAI-compatible endpoint, Ollama and llama.cpp. A cheap model triages, a strong one reviews. Configuration →

Inspectable prompts

Path-scoped instructions in your repo. nitpick explain-config prints the exact prompt. Instructions →

Related context

Definitions the change uses, attached by default. Callers of what it redefines, opt-in. Related context →

Trust model

Policy from the base revision. By default, a PR cannot change endpoints or credentials. Trust model →

Integrated tools as extra evidence

golangci-lint, ruff, eslint, semgrep and more. Their output is evidence for the model. Analyzers →

Published measurements

Five corpora, no model judge. Every instrument bug logged, one retraction included. Findings →

How a review runs

  1. The change is read from GitHub or a local checkout, and the policy it is reviewed under comes from the base revision, not the branch.
  2. Analyzers that are installed run against the changed lines, isolated from the tree, and their output becomes evidence.
  3. Files are bundled into batches under a token budget, with related context attached: what a changed line calls by default, and who calls what the change redefined when the caller walk is switched on.
  4. Each batch is reviewed by the model the route selects. A triage model merges and filters. An optional expert pass refutes.
  5. The review is posted as inline comments, with a summary that lists every file not reviewed, every analyzer that did not run, and every finding that was discarded and why.

The full walkthrough →

The quickstart uses OpenRouter with GLM-5.3-Flash, the reviewer this repository uses in CI. You can spend nothing first: nitpick explain-config prints what a review would send without sending it, and LLM_PROVIDER=ollama runs against a local model. Why, and the alternatives →