Strix runs autonomous AI agents that test an application for security vulnerabilities the way a penetration tester would. The agents run the code dynamically, probe it, and confirm each finding with a working proof-of-concept exploit instead of reporting unverified matches. It targets developers and security teams who want application security testing without a manual pentest engagement or the false positives of static analysis alone.
A scan points at a local codebase, a GitHub repository, a live web application, or an API contract (OpenAPI, Swagger, or a Postman collection) paired with its base URL. Several targets can be combined, for example source code together with the deployed app for gray-box testing. Agents work in a Docker sandbox and organize as a graph of specialized agents for reconnaissance, exploitation, and post-exploitation that share discoveries and chain vulnerabilities.
The open-source CLI runs locally with your own LLM key, routed through LiteLLM. It also runs headless for servers and CI, where a GitHub Actions workflow scans pull requests and scopes quick reviews to changed files. A local web viewer shows findings, a live map of the agent team, and past runs. Strix Cloud adds a hosted platform with continuous scanning, one-click autofix pull requests, and integrations with GitHub, GitLab, Bitbucket, Slack, Jira, and Linear, and an enterprise tier adds SSO, compliance reports, and VPC or self-hosted deployment.
Strix also ships as a set of nine agent skills, so Claude Code, Codex, Cursor, or any SKILL.md-compatible agent can run pentests, fix findings, and set up CI scanning. It can connect to MCP servers and expose their tools to its agents during a run.
Features
- Exploit validation: each vulnerability comes with a working proof-of-concept and reproduction steps
- Multi-agent pentesting: specialized agents for recon, exploitation, and post-exploitation coordinate and run in parallel across targets
- Offensive toolkit: HTTP interception proxy (Caido), browser exploitation, interactive shell, a Python exploit sandbox, and reconnaissance and OSINT
- Vulnerability coverage: access control, injection, SSRF, XXE, deserialization, XSS, business logic flaws, authentication, cloud misconfigurations, and API security
- Static and dynamic analysis: SAST and DAST, with findings scored by CVSS and classified by OWASP category
- API contract testing: tests every endpoint declared in an OpenAPI, Swagger, or Postman definition
- Headless and CI mode: non-interactive runs exit non-zero on findings, and pull request scans cover changed files
- Local web viewer: a dashboard of findings, the agent team, and past runs that keeps data on your machine
- Coding-agent skills: nine skills for running pentests, fixing findings, and CI scanning from Claude Code, Codex, or Cursor
- MCP servers: tools from local or remote MCP servers are exposed to the agents during a run
- Model choice: works with OpenAI, Anthropic, Google, Z.ai GLM, DeepSeek, and Moonshot Kimi models, local models, or a ChatGPT subscription
- Cloud platform: hosted continuous pentesting with autofix pull requests and DevSecOps integrations
