Redefining Software Productivity Beyond Vibe Coding — There’s More Beyond Lovable and Replit.
Moving beyond unreliable prototype generation, Prometheus turns your ideas into verifiable, affordable software through autonomous code agents.
- [2025-11] Powered by GPT-5, Prometheus achieves state-of-the-art performance with 74.4% and 33.8% resolution rates on SWE-bench Verified leaderboard and SWE-Polybench Verified leaderboard, ranking Top-6 and Top-1🏆 among open-source agent systems, respectively🎉!
Prometheus is a research-backed, production-ready platform that leverages unified knowledge graphs and multi-agent systems to perform intelligent operations on multilingual codebases. Built on LangGraph state machines, it orchestrates specialized AI agents to automatically classify issues, reproduce bugs, retrieve relevant context, and generate validated patches.
- Automated Issue Resolution: End-to-end bug fixing with reproduction, patch generation, and multi-level validation
- Feature Implementation Pipeline: Context-aware feature request analysis, implementation planning, and code generation with optional regression testing
- Intelligent Context Retrieval: Graph-based semantic search over codebase structure, AST, and documentation
- Multi-Agent Orchestration: Coordinated workflow between classification, reproduction, and resolution agents
- Knowledge Graph Integration: Neo4j-powered unified representation of code structure and semantics
- Containerized Execution: Docker-isolated testing and validation environment
- Question Answering: Natural language queries with tool-augmented LLM agents
| System | Core Description | Limitations | Why Prometheus is Superior |
|---|---|---|---|
| SWE-Agent | Research baseline for automatic bug fixing using a single LLM-computer interface | – Static, single-turn reasoning – No cross-file or cross-repo understanding – Lacks full detect–reproduce–repair–verify (DRRV) automation |
✅ Prometheus performs multi-agent collaborative reasoning across files and commits, enabling full-cycle issue understanding and repair |
| Lingxi | Multi-agent system for automated bug fixing and code reasoning | – Limited context retrieval – No persistent knowledge graph or long-term memory – Requires human validation for many patches |
✅ Prometheus integrates a Unified Codebase Knowledge Graph and long-term memory (Athena) for deeper semantic reasoning and repository-wide understanding |
| TRAE | Multi-agent reasoning and tool execution framework | – Focused on task orchestration rather than reasoning depth – No unified memory or structured code representation |
✅ Prometheus emphasizes deep reasoning and knowledge unification, allowing consistent understanding across large and complex repositories |
| OpenHands | General-purpose open-source AI developer using sandbox execution | – Strong executor but weak contextual reasoning – No repository-level semantic linking – Task-by-task operation only |
✅ Prometheus combines contextual understanding and code reasoning, achieving coherent, reproducible debugging and intelligent code repair |
Prometheus implements a hierarchical multi-agent system:
User Issue
|
v
┌─────────────────────────────────┐
│ Issue Classification Agent │
│ (bug/question/feature/doc) │
└─────────────┬───────────────────┘
|
┌───────────────┼───────────────┐
| | |
v v v
┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│Bug Pipeline │ │Feature │ │Question │
│ │ │Pipeline │ │Pipeline │
└──────┬───────┘ └──────┬───────┘ └──────┬───────┘
| | |
v v v
┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│Bug │ │Feature │ │Context │
│Reproduction │ │Analysis │ │Retrieval │
└──────┬───────┘ └──────┬───────┘ └──────┬───────┘
| | |
v v v
┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│Issue │ │Patch │ │Question │
│Resolution │ │Generation │ │Analysis │
└──────┬───────┘ └──────┬───────┘ └──────┬───────┘
| | |
└─────────────────┼─────────────────┘
v
Response Generation
Core Components:
- Knowledge Graph: Tree-sitter-based AST and semantic code representation in Neo4j
- LangGraph State Machines: Coordinated multi-agent workflows with checkpointing
- Docker Containers: Isolated build and test execution environments
- LLM Integration: Multi-tier model strategy (GPT-4, Claude, Gemini support)
See Architecture Documentation for details.
- Docker and Docker Compose
- Python 3.11+ (for local development)
- API Keys: OpenAI, Anthropic, or Google Gemini
-
Clone the repository
git clone https://github.com/EuniAI/Prometheus.git cd Prometheus -
Configure environment
cp example.env .env # Edit .env with your API keys -
Generate JWT secret (required for authentication)
python -m prometheus.script.generate_jwt_token # Copy output to .env as PROMETHEUS_JWT_SECRET_KEY -
Create working directory
mkdir -p working_dir
-
Start services
docker-compose up --build
-
Access the platform
- API: http://localhost:9002/v1.2
- Interactive Docs: http://localhost:9002/docs
# Install dependencies
pip install hatchling
pip install .
pip install .[test]
# Run development server
uvicorn prometheus.app.main:app --host 0.0.0.0 --port 9002 --reload# Run tests (excluding git-dependent tests)
coverage run --source=prometheus -m pytest -v -s -m "not git"
# Generate coverage report
coverage report -m
coverage html
open htmlcov/index.htmlPostgreSQL (required for state checkpointing):
docker run -d \
-p 5432:5432 \
-e POSTGRES_USER=postgres \
-e POSTGRES_PASSWORD=password \
-e POSTGRES_DB=postgres \
postgresNeo4j (required for knowledge graph):
docker run -d \
-p 7474:7474 -p 7687:7687 \
-e NEO4J_AUTH=neo4j/password \
-e NEO4J_PLUGINS='["apoc"]' \
-e NEO4J_dbms_memory_heap_initial__size=4G \
-e NEO4J_dbms_memory_heap_max__size=8G \
neo4jVerify at http://localhost:7474
We welcome contributions! Please see our Contributing Guidelines for details on how to get started.
Quick Links:
- 📖 Read the full Contributing Guide
- 🐞 Report bugs via GitHub Issues
- ✨ Submit feature requests and improvements via Pull Requests
We're grateful to all our amazing contributors who have made this project what it is today!
If you have any questions or encounter issues, please feel free to reach out. For quick queries, you can also check our Issues page for common questions and solutions.
This project is dual-licensed:
- Community Edition: Licensed under the GNU General Public License v3.0 (GPLv3).
You are free to use, modify, and redistribute this code, provided that any derivative works are also released under the GPLv3.
This ensures the project remains open and contributions benefit the community.
- Documentation: Multi-Agent Architecture | GitHub Issue Debug Guide
- Issues: GitHub Issues