mantis
A modular, stack-agnostic toolkit for AI coding agents to autonomously find, reproduce, and patch vulnerabilities.
What is mantis?
A modular, stack-agnostic toolkit for AI coding agents to autonomously find, reproduce, and patch vulnerabilities.
Overview of mantis
What it is
Mantis is a modular, stack-agnostic toolkit for AI coding agents to autonomously find, reproduce, and patch vulnerabilities. It is developed by Google and released under the Apache-2.0 license. The toolkit includes a set of skills and an ADK reference harness designed for building secure software. It is intended to be used only in isolated, restricted environments due to the risks of executing autonomously generated code. All findings must be manually verified by a security expert before being reported. The pull request scan feature is untested and should not be used to gate merges.
Key features
- Modular skills for threat modelling, research, review, patching, and reporting that an agent loads on demand.
- An ADK reference harness that runs the skills as a pipeline, with its own two skills to configure and launch it.
- A GitHub Action that runs the harness over a pull request (currently untested end-to-end).
- Supports multiple LLM providers: Vertex AI, Gemini API, Anthropic, OpenAI, and custom OpenAI-compatible endpoints.
- Includes a semantic tree-sitter index with 20+ native grammars across backend, frontend, mobile, smart contracts, shell/IaC/policy, and firmware/RTL.
- Capable of automatic threat model generation, hypothesis research, vulnerability reproduction (static or dynamic), and deduplication of findings.
- Provides a knowledge database (
knowledge.db) for querying accumulated threat models, findings, and prior fixes, with SARIF 2.1.0 export and MCP sidecar support for IDE/coding-agent integration.
Use cases
Mantis is designed for AI coding agents to autonomously find, reproduce, and patch vulnerabilities in software codebases. It can be used to review code history for historical vulnerabilities, build a semantic index of the codebase, generate threat models, and execute research plans. It supports scanning a single file or an entire repository with configurable depth (1 to 100) for fast triage or exhaustive multi-pass audits. It can also be used during development as a sidecar for live diff gating and background scans via MCP.
Getting started
Prerequisites include Linux, WSL2, or macOS (Apple Silicon supports local microVM sandboxes; Intel macOS works in static-only mode; native Windows is unsupported), Python 3.12–3.14 with venv, and LLM credentials (Vertex AI, Gemini API, Anthropic, OpenAI, or custom endpoint).bashcd reference && ./install.shAfter installation, configure and verify:bash.venv/bin/python3 scripts/configure.py --auto.venv/bin/python3 scripts/configure.py --test --probeRun your first scan:bash./run.sh /path/to/repoFor full details, see the repository README.
License and pricing
Mantis is open source under the Apache-2.0 license. Pricing is not stated beyond being open source.
Based on the project's public README and repository metadata.