T3MP3ST

T3MP3ST is a multi-agent offensive-security framework for authorized penetration testing, security research, and education. It uses the AI coding agent already running on your machine as the executor: you point it at a target you are authorized to test and a kill chain runs through recon, exploit, and report stages, driven from a browser War Room or the command line. The project is explicit that it is for systems you own or have written permission to test.

It works keyless by connecting a local coding agent such as Claude Code, Codex, Hermes, OpenCode, or Oh My Pi, so no new API keys or cloud account are required. Alternatively it runs fully offline against a local model server (Ollama by default, or any OpenAI-compatible server such as LM Studio or vLLM), or with a hosted provider key. Tool calling is driven over text, so models without native function calling still work.

The kill chain is organized as eight operators. The recon engine is tool-backed, driving scanners and fingerprinting so each finding traces to real tool output, and downstream operators (Exploiter, Infiltrator, Exfiltrator, Ghost) run the same ReAct loop. An egress-scope containment default refuses off-scope public hosts once a mission target is set. Capabilities are exposed as an Arsenal of built-in tools, over an MCP server, and through an HTTP API, with dangerous drivers gated behind narrow approved paths.

Target domains include web applications, capture-the-flag challenges, source-code analysis across several languages, smart contracts, cloud infrastructure-as-code, mobile, and binary reverse engineering, with a status table marking each as stable, experimental, or roadmap. Every headline number in the project recomputes from committed benchmark data with a single verification command.