PentAGI introduces an AI-driven approach to penetration testing, automating complex workflows with tools like Nmap and Metasploit while generating detailed reports.
Developed by VXControl and released on GitHub in early 2025, this open-source platform empowers security professionals to conduct autonomous assessments in isolated Docker environments.
The tool stands out for its fully autonomous AI agents that dynamically plan and execute pentests, integrating over 20 professional security tools, including Nmap for network discovery, Metasploit for exploitation, and sqlmap for database attacks.
Users define a target, and PentAGI’s multi-agent system, comprising researcher, developer, and executor roles, orchestrates the process, leveraging long-term memory to recall past successes and adapt strategies.
This eliminates manual scripting, enabling rapid vulnerability identification and proof-of-concept exploits without compromising host systems, as all operations run in a sandbox.
PentAGI’s intelligence stems from integrations with leading LLMs like OpenAI, Anthropic Claude, Google Gemini, and local Ollama models, allowing flexible deployment from cloud APIs to on-premises inference.
External search APIs such as Tavily, Perplexity, and DuckDuckGo provide real-time web intelligence, while a built-in scraper gathers target-specific data securely.
The system produces comprehensive reports with exploitation guides, stored persistently in PostgreSQL with pgvector for semantic querying, and visualized via Grafana dashboards for monitoring agent performance.
A sophisticated chain summarization mechanism prevents LLM context overflow, preserving critical conversation history through configurable QA pairs and byte-limited sections. This ensures coherent multi-turn reasoning even in extended pentests.
| Parameter | Environment Variable | Default | Description |
|---|---|---|---|
| Preserve Last | SUMMARIZER_PRESERVE_LAST | true | Keep last section messages intact |
| Last Section Size | SUMMARIZER_LAST_SEC_BYTES | 51200 | Max bytes for last section (50KB) |
| Max QA Size | SUMMARIZER_MAX_QA_BYTES | 65536 | Max bytes for QA sections (64KB) |
Assistant-specific settings allocate more context (up to 75KB), optimizing for complex exploit chains.
At its core, PentAGI employs a microservices architecture with a React/TypeScript frontend, Go-based REST/GraphQL backend, and async task queues for scalability.
Knowledge graphs via Neo4j and Graphiti track entity relationships, enhancing contextual understanding of vulnerabilities. Monitoring stacks like OpenTelemetry, Jaeger, Loki, and VictoriaMetrics provide end-to-end observability, while Langfuse analyzes LLM traces.
Deployment is streamlined via Docker Compose: clone the repo, configure .env with API keys, and launch with a single command, accessible at localhost:8443.
Production setups support horizontal scaling, OAuth (GitHub/Google), and worker nodes for air-gapped execution. Security features include network isolation, TLS, and proxy support for LLM/search traffic.
As AI pentesting evolves, PentAGI addresses key pain points like tool chaining and report automation, positioning it among the top open-source tools for 2026. Security teams can self-host for data control, though users must manage LLM costs and rate limits, especially on AWS Bedrock.
Follow us on Google News, LinkedIn, and X for daily cybersecurity updates. Contact us to feature your stories.
