This post goes over how to use AI (LLMs) to summarize pull requests (PRs) with GitHub Actions.

Action

The easiest method is to use the action pull-request-summary:

# .github/workflows/pull-request-summary.yml
on: pull_request
jobs:
  pull-request-summary:
    runs-on: ubuntu-latest
    permissions:
      contents: read
      pull-requests: write
    steps:
      - uses: ai-action/pull-request-summary@v1

The action reviews the code changes with a large language model (LLM) and edits the pull request description with the response.

See example.

Custom

To create a custom workflow, set up Ollama and checkout the repository:

- uses: ai-action/setup-ollama@v1
- uses: actions/checkout@v4

Then run a prompt with the PR diff against codellama and edit the PR description with the response:

- run: |
    PROMPT=$(printf 'Summarize the following code diff:\n%s' $(gh pr diff $PR_NUMBER))
    LLM='codellama'
    PR_SUMMARY=$(ollama run $LLM "$PROMPT")
    gh pr edit $PR_NUMBER --body "$PR_SUMMARY"
  env:
    GITHUB_TOKEN: ${{ github.token }}
    PR_NUMBER: ${{ github.event.pull_request.number }}

Here’s the full workflow:

# .github/workflows/pull-request-summary.yml
on: pull_request

jobs:
  pull-request-summary:
    runs-on: ubuntu-latest
    permissions:
      contents: read
      pull-requests: write

    steps:
      - uses: ai-action/setup-ollama@v1
      - uses: actions/checkout@v4
      - run: |
          PROMPT=$(printf 'Summarize code diff below:\n%s' $(gh pr diff $PR_NUMBER))
          LLM='codellama'
          PR_SUMMARY=$(ollama run $LLM "$PROMPT")
          gh pr edit $PR_NUMBER --body "$PR_SUMMARY"
        env:
          GITHUB_TOKEN: ${{ github.token }}
          PR_NUMBER: ${{ github.event.pull_request.number }}

See example.