AutoPR-AI

Self-improving PR intelligence platform. Reviews pull requests, generates descriptions, scores codebase health, and learns from your team's engineering patterns.

Architecture

AutoPR-AI/
├── app.py                    FastAPI entry point
├── config.py                 Centralized settings (Pydantic)
├── webhooks.py               GitHub App webhook handler
│
├── core/
│   ├── db.py                 Async SQLAlchemy (SQLite/PostgreSQL)
│   ├── llm.py                Unified LLM client (5 providers)
│   ├── github_app.py         GitHub App JWT auth + API client
│   └── models.py             Pydantic request/response contracts
│
├── modules/
│   ├── pr_description/       Phase 1 — PR Description as a Service
│   │   ├── analyzer.py       Diff + context extraction
│   │   ├── generator.py      Structured description generation
│   │   ├── learner.py        Team format learning
│   │   └── routes.py         /api/describe endpoints
│   │
│   ├── pr_review/            Phase 2 — Self-Improving Review Agent
│   │   ├── reviewer.py       Full-context PR reviewer
│   │   ├── team_dna.py       Team pattern extraction
│   │   ├── feedback.py       Accept/reject tracking + prompt tuning
│   │   └── routes.py         /api/review endpoints
│   │
│   ├── code_health/          Phase 3 — Codebase Credit Score
│   │   ├── scanner.py        5-dimension code scanner
│   │   ├── scorer.py         Weighted 0-100 scoring
│   │   ├── reporter.py       PR comments + GitHub status checks
│   │   └── routes.py         /api/health endpoints
│   │
│   └── refactor/             Legacy — AI Code Refactoring
│       ├── agent.py          Clone → refactor → PR pipeline
│       └── routes.py         POST /refactor endpoint
│
└── tests/
    ├── test_pr_description.py
    ├── test_pr_review.py
    └── test_code_health.py

Modules

PR Description Service

Generates structured PR descriptions from diff context. Reads the diff, commit messages, and linked issues. Outputs: Summary, Why, What Changed, How to Test, Risks. Learns your team's preferred format from historical PRs.

Self-Improving PR Review Agent

Reviews PRs with full codebase context, not just the diff. Builds a Team DNA model from merged PRs — naming conventions, architectural patterns, review culture. Tracks which suggestions get accepted vs rejected and self-tunes prompting based on feedback.

CodeHealth Score

Scans a repository across 5 dimensions: cyclomatic complexity, security vulnerabilities, dependency health, code smells, and documentation coverage. Produces a weighted 0-100 score with letter grades. Updates on every PR merge. Posts score impact as a GitHub status check.

Code Refactor

Points at a GitHub repo, clones it, scans for code smells, refactors with AI, and opens a PR with clean code. Supports 5 LLM providers with free models available.

Webhook Integration

When installed as a GitHub App, AutoPR-AI reacts to events automatically:

• pull_request.opened → generates description + runs review + health scan
• pull_request.synchronize → re-runs health scan on force-push
• pull_request_review_comment → detects accept/reject feedback, feeds learning loop

Setup

git clone https://github.com/ixchio/AutoPR-AI && cd AutoPR-AI
python3 -m venv venv
./venv/bin/pip install -r requirements.txt
cp .env.example .env  # fill in your keys
./venv/bin/python -m uvicorn app:app --host 0.0.0.0 --port 8080

Configuration

Copy .env.example to .env. Required:

Variable	Purpose
GITHUB_TOKEN	Personal access token for GitHub API
OPENROUTER_API_KEY	LLM provider key (at least one required)
DATABASE_URL	sqlite+aiosqlite:///autopr.db (default)

Optional (for GitHub App mode):

Variable	Purpose
GITHUB_APP_ID	GitHub App installation ID
GITHUB_APP_PRIVATE_KEY	GitHub App private key (PEM)
GITHUB_WEBHOOK_SECRET	Webhook signature verification

API

Interactive docs available at /docs when the server is running.

Endpoint	Method	Description
/api/describe/	POST	Generate PR description
/api/describe/learn	POST	Learn team format
/api/review/	POST	Run AI code review
/api/review/build-dna/{owner}/{repo}	POST	Build Team DNA model
/api/review/feedback	POST	Record feedback
/api/health/scan	POST	Scan codebase health
/api/health/trend/{owner}/{repo}	GET	Health score trend
/refactor	POST	Trigger code refactoring
/webhooks/github	POST	GitHub App webhook receiver

LLM Providers

Provider	Env Variable	Free Tier
OpenRouter	OPENROUTER_API_KEY	Yes
OpenAI	OPENAI_API_KEY	No
Anthropic	ANTHROPIC_API_KEY	No
DeepSeek	DEEPSEEK_API_KEY	No
Grok (xAI)	XAI_API_KEY	No

Tests

./venv/bin/python -m pytest tests/ -v

14 tests covering PR description parsing, review output parsing, feedback guidance, and health scoring.

Stack

• Runtime: Python 3.12, FastAPI, uvicorn
• Database: SQLAlchemy async (SQLite default, PostgreSQL ready)
• Auth: PyJWT for GitHub App authentication
• Formatting: black + isort