Apppublic

Zoghby/PatchPilotAI

sourceHugging Facemitupdated 5mo agoView on Hugging Face

2likes

App README

PatchPilot AI

PatchPilot AI is a generic agentic code review, debugging, and improvement copilot for tough complex code and simple everyday scripts.

It supports C++, Java, JavaScript, Python, and C# source files. The app analyzes code, detects bug risks, security issues, performance bottlenecks, complexity problems, and generates actionable fix recommendations plus practical test plans.

Features

—Upload C++, Java, JavaScript, Python, C#, text, or markdown files
—Multi-agent workflow:
—Language & Structure Agent
—Bug & Correctness Agent
—Security Agent
—Performance & Complexity Agent
—Refactor Agent
—Test Planner Agent
—Executive Summary Agent
—Works without API keys using deterministic rule-based analysis
—Optional OpenAI-compatible model support for deeper review
—Designed for GPT-5.5 / GPT-5.4 / GPT-5 family style coding models through API keys
—Streamlit UI for local demo or Hugging Face Spaces

Run locally

bash

pip install -r requirements.txt
streamlit run app.py

Optional OpenAI model mode

Set these as environment variables or enter them in the app sidebar:

bash

OPENAI_API_KEY=<your_key>
PATCHPILOT_MODEL=gpt-5.5

The app calls the OpenAI-compatible Chat Completions endpoint. If no API key is provided, PatchPilot still runs with the deterministic fallback analyzer.

AMD Developer Hackathon relevance

PatchPilot AI fits the AI Agents & Agentic Workflows track. It uses multiple specialized code-review agents to automate a real developer workflow.

For AMD Developer Cloud integration, the same app can point to an OpenAI-compatible endpoint hosted on AMD GPU infrastructure using ROCm/vLLM:

bash

PATCHPILOT_LLM_BASE_URL=<AMD-hosted-endpoint>/v1
PATCHPILOT_LLM_API_KEY=<key>
PATCHPILOT_MODEL=<model-name>

License

MIT