ai-mr-reviewer

README

AI MR Reviewer

ai-mr-reviewer is a Go CLI that reviews GitLab merge requests and GitHub pull requests with an AI model and publishes inline feedback back to the code review.

This is a hobby project. It is not production-ready and should be used with that expectation.

It supports:

• GitLab merge requests
• GitHub pull requests
• Ollama
• OpenAI-compatible endpoints
• Anthropic
• GitHub Models / Copilot-compatible inference

What It Does

For each run, the tool:

1. loads repository and provider configuration from environment variables
2. fetches the current MR/PR diff from GitLab or GitHub
3. combines the diff into a single review prompt
4. sends that prompt to the configured AI provider
5. parses the model response into review issues
6. posts inline comments back to the MR/PR

It can also delete earlier unresolved bot comments before publishing a fresh review pass.

Requirements

• Go 1.26+
• access to either GitLab or GitHub APIs
• one configured AI provider:
  • Ollama
  • OpenAI
  • Anthropic
  • Copilot / GitHub Models

Quick Start

GitHub + OpenAI

export VCS_PROVIDER="github"
export GITHUB_TOKEN="your-github-token"
export GITHUB_OWNER="your-org-or-user"
export GITHUB_REPO="your-repo"
export GITHUB_PR_NUMBER="123"
export CI_COMMIT_SHA="abc123"

export AI_PROVIDER="openai"
export OPENAI_API_KEY="your-openai-key"

go run ./main.go

GitLab + Ollama

export VCS_PROVIDER="gitlab"
export GITLAB_TOKEN="your-gitlab-token"
export CI_PROJECT_ID="123"
export CI_MERGE_REQUEST_IID="1"
export CI_COMMIT_SHA="abc123"
export CI_MERGE_REQUEST_DIFF_BASE_SHA="def456"

export AI_PROVIDER="ollama"
export OLLAMA_URL="http://localhost:11434"
export OLLAMA_MODEL="llama3.2"

go run ./main.go

How Reviews Work

• The reviewer sends one combined diff per run, not one request per file.
• Existing comments are checked so already-reviewed files can be skipped.
• The AI response is expected to contain JSON with an issues array.
• Each issue becomes an inline discussion if the target platform accepts the location.
• RUN_TIMEOUT limits the full run, including API calls and model inference.

Configuration

The application is configured entirely through environment variables.

Core Settings

Variable	Description	Default
VCS_PROVIDER	Review source platform: gitlab or github	gitlab
AI_PROVIDER	AI backend: ollama, openai, anthropic, or copilot	ollama
COMMENT_PREFIX	Prefix added to every bot comment as <prefix>:	ai-mr-reviewer
DELETE_BOT_COMMENTS	Delete previous unresolved bot comments before reviewing	true
RUN_TIMEOUT	Maximum duration for a full review run	10m

GitLab Settings

Variable	Description	Default
GITLAB_TOKEN	GitLab API token	-
CI_PROJECT_ID	GitLab project ID	-
CI_MERGE_REQUEST_IID	Merge request IID	-
CI_COMMIT_SHA	Commit SHA for the review target	-
CI_MERGE_REQUEST_DIFF_BASE_SHA	Base SHA for the MR diff	-
CI_SERVER_URL	GitLab server URL	https://gitlab.com

GitHub Settings

Variable	Description	Default
GITHUB_TOKEN	GitHub token for PR access and, for copilot, model access	-
GITHUB_OWNER	Repository owner	-
GITHUB_REPO	Repository name	-
GITHUB_PR_NUMBER	Pull request number	-
CI_COMMIT_SHA	Commit SHA for the review target	-

Ollama Settings

Variable	Description	Default
OLLAMA_URL	Ollama base URL	http://localhost:11434
OLLAMA_API_KEY	Optional API key for protected or hosted Ollama endpoints	-
OLLAMA_MODEL	Ollama model name	llama3.2

OpenAI Settings

Variable	Description	Default
OPENAI_API_KEY	OpenAI API key	-
OPENAI_BASE_URL	OpenAI-compatible API base URL	https://api.openai.com/v1
OPENAI_MODEL	Model name	GPT-5.2-Codex

Anthropic Settings

Variable	Description	Default
ANTHROPIC_AUTH_TOKEN	Anthropic API token	-
ANTHROPIC_BASE_URL	Anthropic API base URL	https://api.anthropic.com/v1/
ANTHROPIC_MODEL	Model name	claude-sonnet-4-20250514

Copilot / GitHub Models Settings

Variable	Description	Default
GITHUB_TOKEN	GitHub token used as bearer auth for GitHub Models	-
COPILOT_BASE_URL	GitHub Models-compatible inference URL	https://models.github.ai/inference
COPILOT_MODEL	GitHub Models-compatible model name	openai/gpt-4.1

Local Usage Examples

GitLab + Anthropic

export VCS_PROVIDER="gitlab"
export GITLAB_TOKEN="your-gitlab-token"
export CI_PROJECT_ID="123"
export CI_MERGE_REQUEST_IID="1"
export CI_COMMIT_SHA="abc123"
export CI_MERGE_REQUEST_DIFF_BASE_SHA="def456"

export AI_PROVIDER="anthropic"
export ANTHROPIC_AUTH_TOKEN="your-anthropic-token"

go run ./main.go

GitHub + Copilot

export VCS_PROVIDER="github"
export GITHUB_TOKEN="your-github-token"
export GITHUB_OWNER="your-org-or-user"
export GITHUB_REPO="your-repo"
export GITHUB_PR_NUMBER="123"
export CI_COMMIT_SHA="abc123"

export AI_PROVIDER="copilot"
export COPILOT_BASE_URL="https://models.github.ai/inference" # optional
export COPILOT_MODEL="openai/gpt-4.1"                         # optional

go run ./main.go

CI Integration

GitLab CI

stages:
  - review

ai-code-review:
  stage: review
  image: golang:1.26-alpine
  services:
    - name: ollama/ollama:latest
      alias: ollama
  variables:
    VCS_PROVIDER: "gitlab"
    CI_SERVER_URL: $CI_SERVER_URL
    GITLAB_TOKEN: $GITLAB_TOKEN
    CI_PROJECT_ID: $CI_PROJECT_ID
    CI_MERGE_REQUEST_IID: $CI_MERGE_REQUEST_IID
    CI_COMMIT_SHA: $CI_COMMIT_SHA
    CI_MERGE_REQUEST_DIFF_BASE_SHA: $CI_MERGE_REQUEST_DIFF_BASE_SHA
    AI_PROVIDER: "ollama"
    OLLAMA_URL: "http://ollama:11434"
    OLLAMA_MODEL: "llama3.2"
  before_script:
    - go install github.com/mistername/ai-mr-reviewer@latest
  script:
    - ai-mr-reviewer
  rules:
    - if: $CI_PIPELINE_SOURCE == "merge_request_event"
    - if: $CI_MERGE_REQUEST_IID
  allow_failure: true

GitHub Actions

Base workflow:

name: AI Code Review

on:
  pull_request:

jobs:
  review:
    runs-on: ubuntu-latest
    permissions:
      contents: read
      pull-requests: write
      models: read
    steps:
      - uses: actions/checkout@v6
      - name: Set up Go
        uses: actions/setup-go@v6
        with:
          go-version: "1.26"
      - name: Install and run reviewer
        env:
          VCS_PROVIDER: github
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          GITHUB_OWNER: ${{ github.repository_owner }}
          GITHUB_REPO: ${{ github.event.repository.name }}
          GITHUB_PR_NUMBER: ${{ github.event.pull_request.number }}
          CI_COMMIT_SHA: ${{ github.sha }}
          DELETE_BOT_COMMENTS: true
        run: go install github.com/mistername/ai-mr-reviewer@latest && ai-mr-reviewer

Provider-specific additions:

OpenAI:

          AI_PROVIDER: openai
          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
          OPENAI_MODEL: GPT-5.2-Codex

Anthropic:

          AI_PROVIDER: anthropic
          ANTHROPIC_AUTH_TOKEN: ${{ secrets.ANTHROPIC_AUTH_TOKEN }}
          ANTHROPIC_MODEL: claude-sonnet-4-20250514

Ollama:

          AI_PROVIDER: ollama
          OLLAMA_URL: http://localhost:11434
          OLLAMA_API_KEY: ${{ secrets.OLLAMA_API_KEY }} # optional
          OLLAMA_MODEL: llama3.2

If you want GitHub Actions to start an Ollama model locally, add:

      - name: Run model
        uses: ai-action/ollama-action@v2
        with:
          model: llama3.2

Copilot / GitHub Models:

          AI_PROVIDER: copilot
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          COPILOT_BASE_URL: https://models.github.ai/inference
          COPILOT_MODEL: openai/gpt-4.1

models: read is only required when using AI_PROVIDER=copilot, but leaving it enabled is harmless if you switch providers later.

Token Setup

GitLab

Create a GitLab token with enough scope to read merge request data and write discussions.

Recommended process:

1. Go to GitLab -> Settings -> Access Tokens
2. Create a token with the scopes your instance requires for MR read/write operations
3. Store it as GITLAB_TOKEN

If your GitLab instance distinguishes discussion-writing permissions, make sure the token can post MR discussions.

GitHub

For local runs with a personal token:

1. Go to GitHub -> Settings -> Developer settings -> Personal access tokens
2. Create a token with repository access appropriate for your repo visibility
3. If using AI_PROVIDER=copilot, ensure the token can call GitHub Models
4. Export it as GITHUB_TOKEN

For GitHub Actions:

• use ${{ secrets.GITHUB_TOKEN }} for repository access
• add models: read when using AI_PROVIDER=copilot

Development

Build

go install .

Run

go run ./main.go

Test

go test ./...

Single package:

go test ./internal/application

Single test:

go test ./internal/application -run '^TestRunReviewsOnlyNewDiffs$'

Race-enabled suite:

go test -v -race ./...

Lint

golangci-lint run ./...

Generate mocks

go generate ./...

Architecture

The project follows a ports-and-adapters style:

main.go
internal/
  application/   # review workflow orchestration
  domain/        # entities, prompts, and port interfaces
  adapters/
    ai/          # Ollama, OpenAI, Anthropic, Copilot adapters
    config/      # env-based configuration
    github/      # GitHub PR adapter
    gitlab/      # GitLab MR adapter
  testutil/      # test helpers

Important files:

• main.go - Fx wiring and startup flow
• internal/application/reviewer.go - main review orchestration
• internal/domain/use_cases.go - domain port definitions
• internal/adapters/config/config.go - environment-backed config

Notes and Limitations

• Reviews are sent as one combined diff, so very large PRs or MRs can hit model context limits.
• On GitHub, inline comments can fall back depending on review API line-placement rules.
• The tool expects the AI response to be parseable into the expected JSON issue format.
• Unsupported VCS providers or AI providers fail early at startup.

License

MIT