DeepSweep

deepsweep-ai.deepsweep

The Agent Evidence Layer for AI coding agents — a flight recorder and circuit breaker for your workflow. Your AI agents have shell access, repo write permissions, and your credentials. Do you actually know what they're doing? DeepSweep is sub-second, local-first agent security that reviews what your agents can reach and records what it found, in a hash-chained local log you can export. Run an AI code review free and locally, authorize what each agent should do, and catch what it gets wrong the moment it lands: prompt injection attacks & jailbreaks · rules file backdoors & malicious commits · MCP misconfigurations · hardcoded secrets leakage · hallucinated npm/pip packages in your dependencies. DeepSweep integrates natively across your entire agentic toolchain — Agentic IDEs & extensions: Antigravity, Trae, Devin Desktop (formerly Windsurf), Cursor, Kiro, Cline; Autopilots & companions: GitHub Copilot, Gemini Code Assist, Tabnine, Amazon Q Developer; Terminal & git-native agents: Claude Code, OpenAI Codex, Aider. Run your first review free — no account required.