knostic/OpenAnt

OpenAnt from Knostic is the leading open source LLM-based vulnerability discovery product, helping defenders proactively find verified security flaws while minimizing both false positives and false negatives. Stage 1 detects. Stage 2 attacks. What survives is real.

OpenAnt is an open-source LLM-based vulnerability discovery and remediation harness (from the company Knostic). It uses large language models to find security flaws in source code — but instead of just flagging suspicious patterns, it runs a two-stage pipeline: first it detects potential vulnerabilities, then it attacks them (adversarially verifies) to confirm they are real. Only flaws that survive this adversarial check are reported as genuine, which cuts down both false positives and false negatives.

It also includes Auto Patcher, a subsystem that generates candidate fixes for known CVEs or scan findingstest hope to make AI-generated patches trustworthy — it plans, applies the patch to a temporary copy, adversarially challenges it, and writes a human-readable Trust Report that explains whether the patch is recommended, with evidence. It never auto-applies patches to your real repo.

The pipeline is model-agnostic: you configure which LLM provider (Anthropic, OpenAI, Google, Bedrock, OpenRouter, Ollama) handles each of seven pipeline phases — e.g., stronger reasoning models for verification, lighter ones for context generation. There's a CLI (openant), a web UI, and support for incremental/diff-based scans so you can scan only what changed in a PR or commit.

Supported languages: Go, Python, C/C++, JavaScript/TypeScript, PHP, Ruby, Zig, Swift, Rust (some listed as beta).

For quick use: build the Go CLI (requires Go 1.25+), run openant setup llm to configure your LLM, then openant init <repo-url> -l go and openant scan --verify. Exit code 1 means vulnerabilities were found; 0 means clean; 2 means scan error (important for CI).

It's designed for defenders — open-source maintainers can scan their own repos (or submit them for free scanning), and it can integrate into CI to catch issues before they're exploited. The project is Apache 2.0 licensed and welcomes contributions.

Related

  • Project
    liu00222/Open-Prompt-Injection

    An open-source toolkit for implementing and evaluating prompt injection attacks and defenses, featuring specialized tools for detecting and localizing malicious prompts.

  • Project
    AISecurityLab/hackagent

    HackAgent is an open‑source Python SDK/CLI that automates red‑team security testing of AI agents. It runs a library of known attacks (prompt injection, jailbreaks, goal hijacking, tool misuse) against agents built with OpenAI, LangChain, LiteLLM, or Google ADK, using LLM‑generated adversarial prompts and a judging LLM to detect failures. Results are stored locally (SQLite) or synced to a cloud dashboard, and the tool can be installed via pip or as a standalone binary.

  • Project
    openhackai/OpenHack

    OpenHack is an open‑source, LLM‑driven security scanner that analyses a codebase through a recon‑→‑hunter‑→‑validator‑→‑verifier pipeline, presenting findings in an interactive terminal UI and optionally verifying exploits in Docker.

  • Project
    Kritt-ai/open-kritt

    open·kritt is an open‑source, self‑hosted platform that orchestrates AI agents (via Codex, Claude Code, Grok Build, etc.) to run custom, prompt‑driven security workflows on code repositories. It builds focused tasks, runs them in parallel Docker containers, validates findings, de‑duplicates and ranks them, and exports a complete report bundle. Designed for bug‑bounty researchers and DevSecOps, it offers a web UI and a headless CLI, supports any model provider, and is released under AGPL‑3.0.

  • Project
    psyray/oasis

    OASIS is a Python CLI tool that uses locally hosted LLMs (via Ollama or any OpenAI‑compatible server) to scan source code for security vulnerabilities. It runs a fast scan model, then deep‑analysis models, validates findings deterministically, caches results, and outputs canonical JSON plus HTML/PDF/Markdown reports. A password‑protected web dashboard lets users explore findings and ask the assistant for re‑investigations. It supports multi‑model runs, suppression registries, diffing against baselines, and CI exit‑code gating.