Vasco0x4/AIDA

AIDA – AI‑Driven Autonomous Pentesting Agent

What it is – AIDA is an open‑source platform that turns any large‑language model (Claude, Gemini, OpenAI, etc.) into an autonomous security‑assessment assistant. The LLM drives a Docker‑based execution environment, runs reconnaissance and exploitation tools, writes Python payloads on the fly, and records every step in a persistent notebook that a human reviewer can later audit.

Key components

• LLM‑backed agent – model‑agnostic; works with any LLM that supports tool‑calling. The agent decides which actions to take (scan, enumerate, exploit, log).
• Pentest container – a pre‑built Docker image (aida-pentest) containing common tools (nmap, sqlmap, ffuf, nuclei, etc.). Missing tools can be installed at runtime, and the container can be swapped for the much larger Exegol image.
• Tool‑calling API – built on Anthropic’s Model Context Protocol (MCP). The LLM invokes functions such as execute(), scan(), http_request(), python_exec(), and add_card() to interact with the target and record findings.
• Dashboard & notebook – a web UI (default http://localhost:31337) shows live progress, a timeline of attacks, and a searchable list of findings. Findings are automatically scored with CVSS 4.0 and can be exported as PDF reports.
• Authentication & notifications – JWT‑based user roles, plus optional Telegram/Slack/Email alerts for new findings.

What it does

1. Scope definition – you provide a target URL, sub‑domain list, and any exclusions.
2. Autonomous execution – the LLM iteratively runs reconnaissance (sub‑domain enumeration, SSL analysis, tech fingerprinting), vulnerability scanning, and custom exploitation (Python scripts, raw HTTP manipulation).
3. Evidence collection – every command, its raw output, and the reasoning that led to it are stored in the notebook.
4. Human hand‑off – after the run, you review the logged findings, reproduce steps if needed, and generate a final report.

Quick‑start (from the README)

# Clone and start the UI (Docker required)
git clone https://github.com/Vasco0x4/AIDA.git && cd AIDA
./start.sh               # UI on http://localhost:31337

# Launch the agent (defaults to Claude Code)
python3 aida.py --assessment "target-corp"

Optional flags let you pick a specific model (--cli codex), run without prompts (--yes), or expose the UI on a LAN/public domain.

Supported models & clients

Client	How to use
Claude Code	python3 aida.py (auto‑detect)
OpenAI Codex CLI	python3 aida.py --cli codex
Kimi, Gemini, other OpenAI‑compatible APIs	same --cli or --base‑url options
Desktop clients (Claude, ChatGPT)	configured via MCP files

Built‑in agent tools

• execute() – run any shell command in the pentest container.
• scan() – wrappers around nmap, ffuf, nikto, dirb, etc.
• subdomain_enum() – discover sub‑domains.
• ssl_analysis() – TLS/SSL checks.
• tech_detection() – fingerprint web technologies.
• python_exec() – execute arbitrary Python code inside the container.
• http_request() – fully controllable HTTP client with credential placeholders.
• add_card() – log a vulnerability (auto‑calculates CVSS 4.0).
• credentials_add() – store secrets for auto‑injection.

Real‑world impact – The README lists several CVEs discovered by AIDA (e.g., a critical RCE in kestra-io/kestra that made it into the CVE database). This demonstrates that the tool can produce publishable findings.

Typical use cases

• Red‑team / pentest automation – run a first‑pass assessment to surface low‑hanging vulnerabilities before a manual audit.
• Bug‑bounty scouting – quickly enumerate and test a target’s attack surface.
• Continuous security testing – integrate the CLI into CI pipelines to run periodic scans.
• Learning / research – experiment with LLM‑driven exploitation without writing the tooling yourself.

Installation notes

• Requires Docker Desktop and an LLM client that supports tool‑calling.
• The start.sh script can run in development mode (--dev), LAN mode (--lan), or public mode with Let’s Encrypt (--domain).
• A “localhost mode” (--localhost) lets the agent execute commands directly on the host instead of inside the container (useful for debugging).

License – AGPL‑v3, so any modifications must be shared under the same license.

Community – Discord channel for support, plus links to issue tracker and contribution guidelines.

---

Bottom line – AIDA is a genuine, open‑source AI‑agent project that combines LLM reasoning with a full pentesting toolchain. It is aimed at security professionals who want an autonomous assistant to handle the repetitive parts of a penetration test while still keeping a human in the loop for validation and reporting.