PentesterFlow: Open-Source AI Agent Automates Penetration Testing Workflows
PentesterFlow, a new open-source AI tool, aims to automate penetration testing and bug hunting workflows from reconnaissance to reporting, emphasizing human oversight and evidence-based findings.
A new open-source, human-in-the-loop agentic AI command-line tool named PentesterFlow has been released, designed to streamline the entire penetration testing and bug hunting process. Developed to address common shortcomings in existing AI security tools, such as hallucinated findings, poor context retention, and integration issues, PentesterFlow focuses on automating workflows from initial reconnaissance through to final reporting, while crucially maintaining analyst oversight.
The tool's architecture incorporates built-in penetration testing skills, a mechanism for confirming findings with concrete evidence, and a continuous local learning system. PentesterFlow connects to various local or hosted large language models (LLMs), plans actions against a defined target scope, and executes real offensive security tools. A key safety feature is its requirement for explicit analyst approval before executing any sensitive commands, ensuring that the AI operates under human supervision and does not pose a risk to production systems.
PentesterFlow is engineered to assist throughout the complete penetration testing lifecycle. This includes initial scoping, detailed reconnaissance, enumeration of services and vulnerabilities, validation of identified weaknesses, tracking of testing coverage, generation of reports, and ongoing learning to improve future engagements. A demonstration showcased the tool's ability to load a specific "webvuln" skill, send HTTP requests to an orders API, and automatically confirm a high-severity Insecure Direct Object Reference (IDOR) vulnerability, documenting the evidence and impact directly into a Markdown file.
The tool supports a wide array of model backends, including Ollama, LM Studio, Kimi, Groq, Gemini, DeepSeek, OpenRouter, and OpenAI-compatible APIs. Its built-in skills cover common attack vectors like web vulnerabilities, Server-Side Request Forgery (SSRF), Server-Side Template Injection (SSTI), JSON Web Tokens (JWT), GraphQL, race conditions, subdomain takeovers, and deserialization flaws. PentesterFlow also integrates with essential tooling such as shell/bash, HTTP clients, Burp Suite, and browser capture utilities, with support for file operations and common command-line utilities like grep.
A standout feature is PentesterFlow's local Continuous Learning System. This system silently stores user preferences, successful workflow patterns, identified coverage gaps, and lessons learned from failed assumptions. This intelligence is saved in project-specific and personal intelligence files, enhancing the AI's performance in subsequent sessions without requiring retraining of the underlying LLM. To maintain simplicity and security, secrets are automatically redacted before writing to disk, and duplicate memories are deduplicated before being fed back to the model.
Security is a primary concern, with PentesterFlow enforcing permission-gated tool execution and blocking potentially catastrophic shell command patterns. Credentials are also redacted during memory compaction and snapshotting. For isolated lab environments, a "YOLO mode" allows for auto-approval of actions. The tool also offers a companion bridge for direct integration with Burp Suite, enabling testers to send captured traffic into the CLI and import confirmed findings back into Burp for a more cohesive workflow between manual and AI-assisted testing.
Installation is designed to be straightforward, with a shell script for macOS/Linux and a PowerShell command for Windows, ensuring the latest standalone binary is downloaded and its integrity verified via SHA-256 checksum. Users can specify versions, opt for local LLMs, or connect to cloud-based providers. Once configured, users can set a target URL and instruct the agent using natural language commands, such as "test the orders API for broken access control."
PentesterFlow's emphasis on transparent, reproducible evidence and analyst-approved actions positions it as a significant development in the growing field of AI-assisted penetration testing. As agentic AI tools continue to proliferate, PentesterFlow offers a compelling option for security teams seeking to leverage AI automation while mitigating the risks associated with fully autonomous agents.