Pentagi

PentAGI is a self hosted, autonomous AI agent for penetration testing. A user defines a security testing task, and the agent analyzes it, chooses an approach and carries out the testing using a terminal, browser, editor and external search, while the user watches progress in real time through a web interface. Results come back as findings and actionable insights. Everything runs in a sandboxed Docker environment that includes more than 200 professional Kali Linux tools. The agent decides its own next step and can look up current information from search engines such as Google, Tavily and Traversaal. A knowledge graph memory built on Graphiti and Neo4j tracks context, and all commands and outputs are stored in PostgreSQL. A REST API and GraphQL interface allow customization, and Langfuse, Grafana and related tools provide observability so the user can monitor the agent while it works. It works with more than twelve model providers, including OpenAI, Anthropic, Google Gemini, AWS Bedrock, DeepSeek, Kimi and Qwen, local models through Ollama, any OpenAI compatible endpoint and a LiteLLM proxy. Community channels are offered on Discord and Telegram. It is aimed at security researchers and ethical hackers who want to automate testing work on their own infrastructure. The vendor frames it as a tool for penetration testing. The site does not publish pricing. The product is run by the user on their own infrastructure, and model costs depend on the provider chosen, with local Ollama models described as zero cost.