PR Code Review Agent (FastAPI · Celery · Postgres · Redis · Agno + Ollama)

An autonomous code-review service for GitHub pull requests.
It fetches PR diffs, runs lightweight static checks, asks a local LLM (Ollama) via Agno to produce a strict JSON review, stores results in Postgres, and exposes a FastAPI API. Work is executed asynchronously using Celery with Redis as broker/result backend.

---

Features

• API
  • POST /analyze-pr → enqueue a PR review
  • GET /status/{task_id} → check status
  • GET /results/{task_id} → fetch structured JSON results
• Async processing with Celery + Redis
• Persistent results in Postgres (JSONB)
• Local LLM review via Agno + Ollama (e.g., llama3.1, qwen2.5)
• Structured logging for debugging & metrics

---

Tech

• FastAPI – lightning-fast web framework
• Agno – agent framework to orchestrate LLMs
• uv – ultra-fast Python package manager & virtualenvs
• Ollama – local LLM runtime (pull models like llama3.1)
• Postgres – relational DB (we store results as JSONB)
• Celery + Redis – task queue + broker/backend

---

Project Layout

gh-code-review-agent/
    app/
        agents/
            code_reviewer.py 
                 Agno agent for reviewing the PR based on the files using Ollama
        controller/
            github_controller.py 
                controller for the endpoint which exposes all the endpoint
        models/
            db_models.py 
                sqlalchemy db model for  Persistence storage
            schema.py 
                pydantic model for Request/Reponse data parsing, validation, and serialization.
        services/
            github_service.py  
                a service class which fetch the files from github using github apis
            static_checks.py
                service to check the fetch files from github PR for static checks like features, bug etc
        tasks/
            celery_app.py
                celery intilaztation class 
            tasks.py 
                background celery task which performs analysis and stores results based on the task id.
        utils/
            auth_dependancy.py
                auth decorater class which checks if the auth header have verifed JWT for secure API endpoints
            db.py
                sqlalchemy db class which creates tables to the postgrace db based on the models defined in models/db_models.py
        config.py
            config class for fetch env varibles
        main.py
            main class and FastAPu entry point for the endooints and other services.
    etc/ (requirement files)
        base.txt
        dev.txt
Dockerfile
.env.example
.gitignore
pyproject.toml
README.md

---

API

POST /github/analyze-pr

Body

{
  "repo_url": "https://github.com/<owner>/<repo>",
  "pr_number": 123,
  "github_token": "optional_token_for_private_repos"
}

Response

{ "task_id": "uuid", "status": "pending" }

GET github/status/{task-id}

{ "task_id": "uuid", "status": "pending|processing|completed|failed", "error": null }

GET github/results/{task-id}

{
  "task_id": "uuid",
  "status": "completed",
  "results": {
    "files": [
      {
        "name": "main.py",
        "issues": [
          {
            "type": "style",
            "line": 15,
            "description": "Line too long",
            "suggestion": "Break line into multiple lines",
            "severity": "low"
          }
        ]
      }
    ],
    "summary": { "total_files": 1, "total_issues": 1, "critical_issues": 0 }
  }
}

Local Setup (uv · Redis · Ollama · Postgres)

• Prereqs

  • Python 3.11+ (3.12 recommended)
  • macOS/Linux/WSL
  • Git

• install uv

  • macOS/linus

    curl -LsSf https://astral.sh/uv/install.sh | sh
    export PATH="$HOME/.local/bin:$PATH"   # ensure in PATH
    uv --version

• create venv

  •   uv venv .venv
      source .venv/bin/activate  

• Install & run Redis

  • macOS (Homebrew)

    brew install redis
    brew services start redis   # or: redis-server

• Install Ollama, pull the model and run

  • install ollama

    curl -fsSL https://ollama.com/install.sh | sh

  • Pull a model (one-time):

    ollama pull llama3.1    # or qwen2.5:14b, deepseek-r1, etc.

  • Run Ollam

    ollama run llama3.1    # or qwen2.5:14b, deepseek-r1, etc.

• Run API & worker

  • API

    uv run uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload
    

    or if already activated the .venv

    uv run fastapi dev     

  • Worker

    uv run celery -A app.celery_app.celery worker -l INFO

    or if already activated the .venv

    celery -A app.tasks.celery_app.celery worker -l INFO

Security

I have secured all API endpoint with JWT auth token which you need to pass in the auth headers. Sample token.

BEARER_SYSTEM_JWT=

for testing purponse you can uncomment this line in main.py #L11

# app = FastAPI(title = get_settings().APP_NAME,dependencies=[Depends(token_required)])
app = FastAPI(title = get_settings().APP_NAME)