GitHub Autonomous AI Agent

An autonomous AI agent that automatically discovers trending GitHub repositories, generates AI answers to community questions (Issues & Discussions), and creates real code patches and opens PRs for solvable issues — all gated behind a human-in-the-loop approval workflow.

---

✨ Features

Feature	Description
🔍 Trend Hunter	Discovers trending repos in 5 languages (Python, JS, TS, Go, Rust) and scores them by priority
💬 Issue Support	Analyzes issues and generates AI answers with RAG context
🗣️ Discussion Support	Automatically generates replies to GitHub Discussions
🔧 Issue Solver	Solvability analysis → real code patch generation → syntax check → Docker sandbox test
🚀 Autonomous PR	Fork → Branch → Commit → PR pipeline
🛡️ Human Approval	All comments and PRs wait for approval before being sent
📚 RAG Pipeline	Clone repo → index into ChromaDB → smart code search in issue context
🐳 Docker Sandbox	Tests AI-generated code in an isolated environment
⚡ Rate Limiting	Automatically manages GitHub API limits

---

🛠️ Architecture

┌──────────────────────────────────────────────┐
│             FastAPI (api/main.py)            │
│  /health  /stats  /pending  /approve  /...   │
└──────────────────────┬───────────────────────┘
                       │
┌──────────────────────▼───────────────────────┐
│              Agent Orchestrator              │
│              6-phase pipeline:               │
│    Trend Hunt → Repo Setup → Community →     │
│  Discussions → Issue Solving → PR Pipeline   │
└─────┬─────────┬──────────┬──────────┬────────┘
      │         │          │          │
┌─────▼────┐ ┌──▼───────┐ ┌▼────────┐ ┌▼───────┐
│  GitHub  │ │   LLM    │ │   RAG   │ │ Docker │
│  Client  │ │Providers │ │ChromaDB │ │Sandbox │
│(GQL+REST)│ │Groq/Olla.│ │         │ │        │
└──────────┘ └──────────┘ └─────────┘ └────────┘

---

🚀 Quick Start

Prerequisites

• Python 3.10+
• PostgreSQL (data storage)
• Ollama + the qwen3-coder:30b model
• Docker (for sandbox tests)
• Git (for cloning repos)

1. Installation

# Virtual environment
python -m venv venv
.\venv\Scripts\Activate.ps1   # Windows
# source venv/bin/activate     # Linux/Mac

# Dependencies
pip install -r requirements.txt

# Ollama model (start.ps1 pulls this automatically; you can also do it manually)
ollama pull qwen3-coder:30b

2. Configuration

copy .env.example .env    # Windows
# cp .env.example .env    # Linux/Mac

Edit the .env file:

• GITHUB_TOKEN → Create a GitHub PAT (required scopes: repo, read:discussion, write:discussion)
• POSTGRES_* → PostgreSQL connection details

3. Database

Schema changes are managed with Alembic (create_all() does not add columns to existing tables).

# Fresh / empty database:
python init_db.py && alembic stamp head

# Update an existing database (apply pending migrations):
alembic upgrade head

# Generate a new migration after changing a model:
alembic revision --autogenerate -m "description" && alembic upgrade head

4. Running

Single entry point — starts everything with one click: PostgreSQL service + Ollama (serve & model pull) + DB init + FastAPI (:8000) + Next.js (:3000).

powershell -ExecutionPolicy Bypass -File .\start.ps1

FastAPI only (e.g. for API debugging):

venv\Scripts\python.exe -m uvicorn api.main:app --host 0.0.0.0 --port 8000

Dashboard: http://localhost:3000 · Swagger UI: http://localhost:8000/docs

---

📡 API Endpoints

Method	Endpoint	Description
GET	/	Agent status
GET	/health	Service health (DB, Ollama, GitHub, Docker)
GET	/agent/stats	Statistics
POST	/agent/trigger?task_type=...	Trigger a task manually
GET	/agent/pending-actions	Code changes awaiting approval
GET	/agent/pending-comments	Comments awaiting approval
POST	/agent/approve-action/{id}	Approve code change → open PR
POST	/agent/reject-action/{id}	Reject code change
POST	/agent/approve-comment/{id}	Approve comment → post to GitHub
POST	/agent/reject-comment/{id}	Reject comment
GET	/agent/actions?limit=20	Action history

---

⚙️ Configuration Options

Variable	Default	Description
OLLAMA_MODEL	qwen3-coder:30b	AI model
TRENDING_DAYS_AGO	7	How many days back to search for repos
MIN_STARS_THRESHOLD	50	Minimum star count
LOOP_INTERVAL_SECONDS	3600	Main loop interval (seconds)
TASK_CONCURRENCY	3	Number of repos processed in parallel
REQUIRE_APPROVAL_FOR_PR	true	Whether PRs require approval
REQUIRE_APPROVAL_FOR_COMMENT	true	Whether comments require approval

---

📁 Project Structure

github-agent/
├── agent/
│   ├── orchestrator.py      # Main orchestrator (6-phase pipeline)
│   ├── ai/                  # AIReasoningService (provider-agnostic reasoning)
│   ├── providers/           # Swappable LLM providers (Groq / Ollama / HF)
│   ├── prompts/             # Jinja2 prompt templates
│   ├── rag/                 # Chunking, retrieval, reranking
│   ├── scoring/             # Repo & issue prioritization
│   ├── solver/              # Agentic code-fix loop (patch + verify)
│   ├── trends/              # Trend aggregation (GitHub, HN, Reddit)
│   └── tools/
│       ├── github_client.py # GitHub API (GraphQL + REST)
│       ├── chroma_client.py # ChromaDB indexing & search
│       └── docker_env.py    # Docker sandbox
├── api/
│   └── main.py              # FastAPI endpoints
├── core/
│   └── config.py            # Pydantic settings (.env)
├── database/
│   ├── models.py            # SQLAlchemy models
│   └── session.py           # DB connection management
├── dashboard/               # Next.js approval & monitoring UI
├── alembic/                 # DB schema migrations
├── tests/                   # Pytest suite
├── .github/workflows/       # CI (lint + typecheck + tests + build)
├── start.ps1                # One-click launcher (single entry point)
├── init_db.py               # DB bootstrap
├── pyproject.toml           # Ruff + pytest config
└── requirements.txt

---

📄 License

Released under the MIT License.

---

Powered by Groq / Ollama (swappable LLM) + GitHub GraphQL API + ChromaDB + Docker