Architecture • Tech Stack • Features • Getting Started • Demo Video • Backend Service
AutoDevOps AI is a next-generation autonomous agent built to audit, diagnose, and stabilize codebases without human intervention. Powered by Google Gemini 3.0, it acts as a virtual Site Reliability Engineer (SRE) that recursively reasons about system architecture, detects high-confidence vulnerabilities, and generates precise, verified fixes.
Modern AI tools assist engineers by suggesting fixes, but stop short of execution, verification, and accountability. In real production incidents, this gap forces humans to manually copy, run, and trust AI-generated code under pressure.
AutoDevOps AI closes this gap by introducing an autonomous agent that not only reasons about failures, but acts, verifies outcomes, and intentionally stops within defined safety boundaries.
AutoDevOps AI acts as an intelligent orchestrator that interfaces with your Git repositories. It uses a hybrid architecture: a React-based Autonomous Agent for reasoning and state management, and a Lightweight Python Microservice for secure file operations.
Each execution occurs in an isolated, ephemeral sandbox; source code is automatically destroyed after analysis, while audit artifacts are preserved.
- Ingest: User provides a repository URL. The agent requests the backend to clone it securely to a temporary sandbox.
- Audit: The agent traverses the file tree and sends high-level context to Gemini 3.0 to identify architectural weaknesses and bugs.
- Reason: Implementing a "Thought Signature" system, the agent maintains a persistent chain of thought across multiple API calls, ensuring consistency.
- Stabilize: For every detected issue, the agent generates a specific, minimal fix strategy, validates it, and prepares a stabilization patch.
- Verify: A multi-path verification step ensures that fixes do not introduce regressions.
- React 19: Utilizing the latest concurrent features for smooth agent state updates.
- Vite: Blazing fast build tool and dev server.
- Google GenAI SDK: Direct integration with Gemini 3.0 models for high-speed inference.
- Firebase: Firestore for persisting agent sessions, logs, and simulation history.
- Tailwind CSS: Rapid, utility-first styling for the mission control dashboard.
- FastAPI: High-performance Python framework for handling Git operations asynchronously.
- Microservice Architecture: Stateless, container-ready design suitable for Cloud Run.
- Git: Native integration for cloning and managing repositories.
- Autonomous Auditing: Executes tests and analyzes failures to identify high-confidence bugs and architectural risks present in the current branch context.
- Recursive Reasoning: Maintains a persistent "Thought Signature" audit log to rationalize decisions, ensuring transparency and consistency across fix attempts.
- Self-Healing: Generates and applies targeted code patches, verifying fixes against the test suite to prevent regressions.
- Live Mission Control: Watch the agent "think" in real-time via a strictly typed event log dashboard.
- Secure Sandboxing: All code operations happen in isolated sessions that are automatically purged.
- Node.js (v18+)
- Python (v3.9+)
- Google Gemini API Key
# Install dependencies
npm install
# Configure Environment
# Copy .env.sample to .env and fill in your credentials
cp .env.sample .env
# Run the Agent Dashboard
npm run devcd backend-simple
# Install Python dependencies
pip install -r requirements.txt
# Start the Microservice
python main.pyThe frontend will automatically discover the local backend at http://localhost:8000.
Distributed under the MIT License. See LICENSE for more information.