Skip to content

All pages

PentAGI

Fully autonomous AI Agents system capable of performing complex penetration testing tasks

AI5d ago
Category
AI
Website
github.com
Language
English
Listed
5d ago
Forks
3,170

vxcontrol/pentagi ↗, Go, MIT, Last commit 8d ago

PentAGI is a self-hosted system of AI agents for automated penetration testing, aimed at information security professionals and researchers. Every operation runs inside an isolated Docker sandbox, and the image is picked automatically to match the task. A built-in toolkit of more than 20 security tools including nmap, metasploit and sqlmap is available to the agents, which are organised as a team: specialised agents handle research, development and infrastructure work under a delegation system with optional execution monitoring and task planning. Findings and working approaches are kept in a long-term memory store in PostgreSQL with pgvector, with an optional Graphiti knowledge graph on Neo4j. Web intelligence comes through a bundled browser and search integrations such as Tavily, Firecrawl, Perplexity, DuckDuckGo and Searxng. The project connects to more than ten LLM providers (OpenAI, Anthropic, Google Gemini, AWS Bedrock, Ollama, DeepSeek and others), exposes REST and GraphQL APIs with bearer tokens, and reports through a web UI plus Grafana and Prometheus integration. It is written in Go with a React front end and released under the MIT license.

Like this product?