Automated code review is one of those ideas that sounds obvious but rarely gets implemented well. The typical approach is to call GPT-4 in a GitHub Action, paste the diff into the prompt, and post the result as a PR comment. It works for small diffs, but it falls apart quickly: large PRs exceed token limits, costs scale linearly with PR size, and you are paying premium prices even for PRs that just rename a variable or update a dependency version.
Antbase makes automated PR reviews economically viable for every PR, not just the ones you manually flag. Small diffs — config changes, version bumps, simple refactors — get routed to fast free models that can still catch obvious issues. Complex diffs with business logic changes get escalated to premium models that understand the nuance. You are not paying $0.10 per review for a one-line change, and you are not skimping on intelligence for a critical security-related PR.
The CI/CD context also benefits from Antbase failover. GitHub Actions have strict timeouts, and if your hardcoded OpenAI call times out or returns a 500, your workflow fails and the review never posts. Antbase retries across providers transparently, so your review workflow is resilient even during provider outages. That matters when you have a team waiting on CI to go green.
GitHub Actions can trigger on pull requests. By calling Antbase with the PR diff, you get automated code review comments — catching bugs, suggesting improvements, and flagging security issues. Antbase routes the review to the best model for code analysis.
Workflow File
# .github/workflows/ai-review.yml
name: AI PR Review
on:
pull_request:
types: [opened, synchronize]
jobs:
review:
runs-on: ubuntu-latest
permissions:
pull-requests: write
steps:
- uses: actions/checkout@v4
with:
fetch-depth: 0
- name: Get PR diff
id: diff
run: |
git diff origin/${{ github.base_ref }}...HEAD > diff.txt
echo "diff<<EOF" >> $GITHUB_OUTPUT
head -c 10000 diff.txt >> $GITHUB_OUTPUT
echo "EOF" >> $GITHUB_OUTPUT
- name: AI Review
env:
ANTBASE_API_KEY: ${{ secrets.ANTBASE_API_KEY }}
run: |
REVIEW=$(curl -s https://antbase.ai/v1/chat/completions \
-H "Authorization: Bearer $ANTBASE_API_KEY" \
-H "Content-Type: application/json" \
-d "$(jq -n --arg diff "${{ steps.diff.outputs.diff }}" '{
model: "auto",
messages: [
{role: "system", content: "You are a code reviewer. Review the diff and provide actionable feedback. Focus on bugs, security issues, and improvements. Be concise."},
{role: "user", content: ("Review this PR diff:\n\n" + $diff)}
]
}')" | jq -r '.choices[0].message.content')
gh pr comment ${{ github.event.number }} --body "## AI Review\n\n$REVIEW"
env:
GH_TOKEN: ${{ github.token }}How It Works
- •On every PR open or push, the workflow extracts the diff and sends it to Antbase.
- •Antbase classifies the request as code analysis and routes to a capable model (typically GPT-4o or Claude for complex diffs).
- •The AI response is posted as a PR comment using the GitHub CLI.
- •The diff is truncated to 10,000 characters to stay within token limits — adjust based on your needs.
Tips
- •Store ANTBASE_API_KEY in GitHub repository secrets (Settings > Secrets and variables > Actions).
- •For large PRs, split the diff by file and make multiple AI calls — one per changed file.
- •Add a label filter to only review PRs with a specific label (e.g., needs-ai-review).
- •Cost is minimal — most PR reviews are a single API call. Antbase may route to a free model for small diffs.
