PentAGI
Self-hosted AI agent system that runs penetration testing tasks in an isolated Docker sandbox.
- Stars
- 33k
- Forks
- 5.2k
- Licence
- MIT
- Language
- Go
- Last update
- 14 days ago
- Type
- AI model / agent
What it is
PentAGI is an open-source, self-hosted system where you describe a target and a goal in plain language, and a team of AI agents plans the work and runs security tools such as nmap, Metasploit and sqlmap inside an isolated Docker sandbox. It keeps results in PostgreSQL with pgvector, and it produces a report at the end. It is meant for authorized security testing on your own infrastructure.
- Describe a target and goal in plain language and let agents plan and run scans
- Store findings and successful approaches in PostgreSQL with pgvector for later runs
- Generate vulnerability reports with exploitation guides
- Track hosts, services and findings in an optional Neo4j knowledge graph
Great for
- Security engineers and penetration testers who want an autonomous assistant on their own servers
- Red teams testing systems they are authorized to assess
- Researchers who want to use local models through Ollama or vLLM
- Teams that want REST or GraphQL API access to automate testing
Not ideal for
- Anyone testing systems without explicit authorization, since the project states it is for authorized testing only
- Beginners with no Docker or security background, because setup and safe use require operational knowledge
- Users who want a fully hosted service with no server to run, since it is self-hosted
How to get it
- 1
See the README
You'll need: Docker · Docker Compose
Clone the repository
git clone https://github.com/vxcontrol/pentagi - 2
Run it
- Clone the repository and follow the Quick Start section of the README
- Run the interactive installer, which sets PentAGI up with Docker Compose
- Configure an LLM provider such as OpenAI, Anthropic or Ollama
- Log in to the web UI and start a task by describing a target and goal
- 3
Get the most out of it
- Use the optional execution monitoring and task planning features when running smaller open models
- Connect the optional Graphiti knowledge graph with Neo4j to track relationships between hosts and findings
- Add search integrations such as Tavily or SearXNG to improve information gathering
- Use the REST or GraphQL APIs with Bearer tokens to automate runs
Downloads
Strengths
- Runs on your own infrastructure with the LLM provider of your choice, including local models
- Tools run inside an isolated Docker sandbox
- Offers REST and GraphQL APIs with Bearer token authentication for automation
Trade-offs
- Requires running Docker and configuring an LLM provider, which takes operational effort
- Autonomous tool use against targets carries real risk and needs careful authorization and scope control
- The README is truncated, so some details of limits and stability could not be verified
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| pentagi (this) | 33k | AI model / agent | advanced | 14 days ago | Security engineers and penetration testers who want an autonomous assistant on their own servers |
| nebula | 18.5k | App | advanced | 7 days ago | Teams connecting servers across several clouds or data centers |
| PentestGPT | 16.2k | AI model / agent | advanced | 4 months ago | Security students working through CTF challenges |
| hermes-agent | 260k | AI model / agent | intermediate | 4 days ago | Developers who want a personal agent they can host on their own server or machine |
| invisible_dots | 32.4k | AI model / agent | intermediate | 6 days ago | Developers who want a self-hosted agent and are comfortable with a command-line setup |
| edict | 17k | AI model / agent | advanced | 1 months ago | Developers already running OpenClaw who want a multi-agent workflow with visible task history |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.