pentagi

Overview of PentAGI: Autonomous AI for Penetration Testing

PentAGI is an autonomous AI-driven system designed for conducting complex penetration testing tasks. Developed by VXControl and released in 2025 under the MIT license, it integrates artificial intelligence with traditional security tools to automate and enhance the efficiency of security assessments. The system is built in Go and is tailored for security professionals, researchers, and enthusiasts who require a scalable, flexible, and isolated environment for testing.

The project emphasizes autonomy, allowing AI agents to determine and execute penetration testing steps independently, with optional human oversight. PentAGI is not a breach and attack simulation (BAS) tool like CALDERA but instead focuses on autonomous, adaptive testing workflows. Its architecture supports horizontal scaling, making it suitable for both small-scale and enterprise-level deployments.

PentAGI operates within a sandboxed Docker environment, ensuring complete isolation for all operations. The tool is self-hosted, providing users with full control over deployment and data management, which is critical for compliance and security-sensitive operations.

How PentAGI Works: Architecture and Execution Flow

The system is divided into distinct components that interact through well-defined APIs, ensuring modularity and scalability.

Core Services

The AI agents operate as a multi-agent system, where each agent is specialized for specific roles such as research, development, or execution. These agents dynamically interact with the system to perform tasks, leveraging a structured flow of commands, subtasks, and actions.

The database stores all commands, outputs, and results, enabling persistent storage and retrieval of testing artifacts. The task queue ensures that operations are processed asynchronously, improving system responsiveness and efficiency. The frontend provides a clean, intuitive interface for managing and monitoring these processes, with real-time updates and detailed logging.

Knowledge Graph and Semantic Relationships

PentAGI integrates Graphiti, a knowledge graph API built on Neo4j, to enable semantic relationship tracking and advanced context understanding. The knowledge graph enhances the AI agents’ ability to make informed decisions by contextualizing data from various sources, including external search systems and internal research.

The knowledge graph is particularly useful for long-term storage of research results and successful approaches, which can be reused in future testing sessions. This capability reduces redundancy and accelerates the testing process by allowing agents to build upon previously validated methodologies.

Monitoring and Observability

These tools provide real-time system observation, including metrics, logs, and distributed tracing. Grafana dashboards offer visual representations of system health, agent performance, and resource utilization, while Prometheus collects and stores time-series data for analysis.

Additionally, PentAGI supports Langfuse for LLM observability, enabling detailed tracking of agent interactions, tool calls, and execution metrics. This integration provides transparency into the decision-making process of the AI agents, which is critical for auditing and improving the reliability of automated testing.

Security Tools and Execution Environment

PentAGI includes a built-in suite of over 20 professional security tools, such as Nmap, Metasploit, and SQLmap, all executed in isolated Docker containers. This sandboxed environment ensures that testing activities do not compromise the host system or other processes. The system automatically selects the appropriate Docker image based on the specific task requirements, optimizing resource usage and security.

A web intelligence module, featuring an isolated browser scraper, enables the system to gather the latest information from web sources dynamically. This capability is essential for reconnaissance phases of penetration testing, where up-to-date data can significantly impact the effectiveness of the assessment.

Use Cases for PentAGI in Security Testing

PentAGI is designed to address a wide range of penetration testing and red teaming scenarios, where automation and intelligence can enhance both speed and accuracy.

Automated Vulnerability Assessment

One of the primary use cases for PentAGI is automated vulnerability assessment. The system can independently scan target systems for vulnerabilities, analyze potential exploitation paths, and generate detailed reports. By integrating professional tools like Nmap and Metasploit, PentAGI automates repetitive tasks such as port scanning, service enumeration, and exploit testing, allowing security teams to focus on strategic analysis and remediation.

This prioritization is enhanced by the knowledge graph, which provides a holistic view of the system’s security posture by mapping relationships between vulnerabilities and system components.

Red Teaming and Adversary Simulation

PentAGI is well-suited for red teaming exercises, where the goal is to simulate real-world cyberattacks to identify weaknesses in an organization’s defenses. The system’s autonomous agents can simulate the behavior of an adversary, from initial reconnaissance to exploitation and post-exploitation activities.

The optional execution monitoring feature allows security teams to oversee the agents’ actions in real time, ensuring that testing remains within ethical and legal boundaries. This oversight is critical for red teaming engagements, where the objective is to identify vulnerabilities without causing unintended damage.

Continuous Security Testing and Compliance

Its REST and GraphQL APIs enable seamless integration with tools such as n8n, OpenClaw, and custom solutions, allowing organizations to embed penetration testing into their development and deployment workflows.

For compliance purposes, PentAGI generates thorough vulnerability reports with exploitation guides, providing clear documentation of findings. The system’s persistent storage ensures that all testing artifacts are retained for future reference and auditing.

Research and Development for Security Tools

PentAGI’s advanced features, such as its knowledge graph and semantic search capabilities, make it a valuable tool for security researchers and developers. The system can be used to explore new attack vectors, test hypotheses about system vulnerabilities, and develop innovative security tools. The integration with external search systems, including Sploitus and Perplexity, enables agents to gather the latest research and exploit databases dynamically.

The delegation system allows researchers to assign specialized tasks to different AI agents, such as infrastructure management or tool development. This modular approach accelerates the research process by parallelizing tasks and leveraging the strengths of each agent type.

Key Features and Capabilities of PentAGI

PentAGI offers a comprehensive set of features that distinguish it as a leading tool for autonomous penetration testing.

• Fully Autonomous AI Agents: PentAGI’s AI agents operate autonomously, determining and executing penetration testing steps without constant human intervention. Optional monitoring allows for oversight and intervention when necessary, ensuring reliability and adaptability.
• Secure and Isolated Execution: All operations are performed in sandboxed Docker containers, providing complete isolation from the host system and other processes.
• Built-in Suite of Professional Tools: The system includes over 20 professional security tools, such as Nmap, Metasploit, and SQLmap, which are essential for comprehensive vulnerability assessment and exploitation testing.
• Smart Memory System: PentAGI stores research results and successful approaches in a persistent PostgreSQL database with pgvector support. This long-term memory enables agents to reuse validated methodologies, improving efficiency and reducing redundancy in testing.
• Knowledge Graph Integration: The optional integration with Graphiti and Neo4j allows PentAGI to track semantic relationships between vulnerabilities, exploits, and system components. This feature enhances context understanding and decision-making during testing.
• Web Intelligence and External Search: PentAGI includes a built-in browser scraper for gathering real-time information from web sources. Additionally, it integrates with advanced search APIs such as Tavily, Firecrawl, and Sploitus, enabling comprehensive information gathering for reconnaissance phases.
• Team of Specialized AI Agents: The system supports a delegation model where tasks are assigned to specialized agents for research, development, and infrastructure management. This modular approach optimizes performance, particularly when using smaller AI models.
• Comprehensive Monitoring and Reporting: PentAGI provides detailed logging and integrates with Grafana and Prometheus for real-time system observation. It generates thorough vulnerability reports with exploitation guides, supporting multiple output formats, including PDF and Markdown.
• Smart Container Management: The system automatically selects the appropriate Docker image based on task requirements, ensuring optimal resource usage and security. This dynamic selection process enhances flexibility and efficiency during testing.
• Modern Interface and APIs: PentAGI features a clean, intuitive web UI for system management and monitoring. It also offers full-featured REST and GraphQL APIs with Bearer token authentication, enabling seamless integration with automation platforms and custom solutions.
• Flexible LLM Provider Support: The system supports over 10 LLM providers, including OpenAI, Anthropic, Google AI (Gemini), AWS Bedrock, Ollama, DeepSeek, GLM, Kimi, Qwen, and MiniMax. It also integrates with aggregators like OpenRouter and DeepInfra, providing broad flexibility in model selection and deployment.
• Quick Deployment and Scalability: PentAGI can be deployed easily using Docker Compose, with comprehensive environment configuration options. Its microservices-based architecture supports horizontal scaling, making it suitable for both small-scale and enterprise deployments.
• These options enhance security, monitoring, and deployment flexibility.

Release v2.1.0 (May 29, 2026)

• User Resources: A persistent, per-user file library with MD5-deduplicated storage, supporting full CRUD operations via REST and GraphQL APIs. Files can be uploaded, managed, and synchronized with worker containers, with protections against directory traversal and symlink escapes.
• Flow Files: Per-flow workspace files that sync into Docker containers, allowing files to be pulled back out of running containers and promoted into the user library. These files are injected into agent system prompts as structured blocks, ensuring agents are aware of user-provided resources.
• Knowledge Base Management: The pgvector memory store is now a first-class, user-manageable resource with semantic search capabilities. Users can perform CRUD operations, semantic searches, and real-time subscriptions, with support for admin and user scoping.
• ToolCall Observability: Real-time logging of tool calls enhances transparency and debugging capabilities, allowing users to monitor agent interactions and execution flows in detail.
• Frontend Modernization: The release includes updates to the frontend, such as React 19, Apollo Client v4, and Vite 8, improving performance and user experience. Stability and security fixes were also addressed in this version.

Release v2.0.0 (April 11, 2026)

This release expanded PentAGI’s LLM provider ecosystem and introduced several key features:

• Each provider is configurable via environment variables and integrated into the GraphQL schema and settings UI, with pre-configured model lists and pricing information.
• Ollama Cloud Support: PentAGI now supports cloud-hosted Ollama models, with pre-built configurations for models such as nemotron-3-super and qwen3-coder-next. Both free and paid tier setups are documented, expanding deployment options for users.
• Analytics Dashboards: A new analytics dashboard was introduced, surfacing usage statistics and cost data collected from REST API endpoints. The dashboard provides token usage breakdowns, cache hit rates, tool call frequency metrics, and per-model cost details, useful for optimizing resource allocation.

Release v1.2.0 (February 25, 2026)

This release focused on enhancing AI capabilities and integration features:

• Support was also added for OpenRouter and OpenAI-compatible endpoints with reasoning content preservation.
• Native caching support for Anthropic and Gemini providers, along with automatic cache hit tracking, was implemented to optimize performance and reduce expenses.
• Usage Analytics and Monitoring: Comprehensive REST API endpoints were added for tracking resource utilization, including token usage breakdowns, cost analysis, execution time metrics, and tool call frequency statistics. These endpoints laid the foundation for the visual analytics dashboard introduced in later releases.
• API Token Management: JWT-based API authentication was implemented, enabling programmatic access to PentAGI’s REST and GraphQL APIs. This feature supports integration with automation platforms such as n8n and OpenClaw, as well as custom solutions, enhancing PentAGI’s versatility in automated workflows.
• This integration enables detailed tracking of agent interactions, tool calls, and execution metrics, improving transparency and debugging capabilities.

Sources

• github.comhttps://github.com/vxcontrol/pentagi