What is Strix?
Strix is an open-source application security testing agent from the company of the same name. You point the strix command at a local code folder, a GitHub repository, a live URL or an API specification such as OpenAPI or a Postman collection, and a coordinating agent splits the work between specialised sub-agents that map the target, test it and write up what they could reproduce, with remediation advice and, for source code, suggested patches. Results are saved under strix_runs/ and can be browsed in a local web viewer.
The agents work inside a Kali Linux-based Docker container with a browser, an HTTP proxy, a terminal and standard security tools. Scans come in quick, standard and deep modes, a headless mode fails a CI job when it finds something, and in pull-request runs it limits quick scans to changed files. Models are reached through LiteLLM, so most hosted providers and local servers work with your own key; a ChatGPT Plus or Pro sign-in is also supported.
The code is Apache-2.0 and written in Python. The company sells a hosted version, Strix Cloud, and an Enterprise edition with VPC or self-hosted deployment.
Is Strix worth it?
Strix gives a security team an open, permissively licensed agent that can test a web app or API from the outside, read its source when you have it, and keep only findings it could reproduce. It has about 75,600 GitHub stars, a little over 15 months after the repository opened in August 2025, and v1.7.1 shipped on 10 November 2026. Unlike Shannon, which needs source code and is AGPL-3.0, Strix also runs black-box against a URL, and Apache-2.0 makes internal use and modification straightforward.
It is not a passive scanner. The agents send real traffic and run tools from their container, and with credentials you supply they test as a logged-in user, so a scan can change data and trip alerts. The README says to run it only against systems you own or have explicit, written permission to test, and to stay within the agreed scope; that means staging copies with disposable data, a separate model key, and a written scope from the system's owner before the first run. The README adds that unauthorised testing is illegal in most jurisdictions and that the software comes with no warranty.
Cost is the model bill, and it is hard to predict. Strix's own benchmark, run on v0.4.0, solved 100 of 104 XBOW challenges in black-box mode for about $337 in total, roughly 19 minutes each; that is the vendor's figure on an old version and a benchmark built from known challenges. An independent black-box test by Protego spent about $17 on Claude Sonnet 4.5 in 10 minutes, sent 5,716,615 input tokens over 105 model requests, and found nothing, because most requests went to routes that did not exist. The tester's advice, which we share, is to give it the source. Deep, the default mode, runs 1 to 4 hours, so set --max-budget on every run.
The catch: data handling needs attention. Source code and app responses go to whichever model provider you choose unless you run a local model, and the docs warn that most local models under 70B parameters struggle with the job. Telemetry to PostHog, Scarf and OpenTelemetry is on by default until you set STRIX_TELEMETRY=0, the API key is saved in ~/.strix/cli-config.json, and the frontier OpenAI, Anthropic and Gemini models need each provider's cyber access programme before they will do security testing. The project moves quickly, with 604 open issues and pull requests.
Our take: think of Strix as a contract tester you can rerun for the price of tokens, useful on staging before each release and on pull requests in quick mode. It does not replace a human pentest for compliance, and a person should confirm every finding before it becomes a ticket.
Where it earns its place
An AppSec lead at a 150-person SaaS company with one annual external pentest
Cover the internal apps that never get a human test between audits
Run standard-mode scans against each app's staging copy, with source attached and --max-budget set per run, under a written scope signed by each app's owner. Confirmed findings go to the security team for review, not straight to developers.
A platform team running GitHub Actions on a payments API
Catch exploitable regressions in pull requests before merge
Add the documented workflow with strix -n -t ./ --scan-mode quick, store STRIX_LLM and LLM_API_KEY as secrets, and point it at a test environment. In CI it scopes quick scans to changed files and exits non-zero when it finds something.
A security engineer at a bank that cannot send code to a hosted model
Test an internal service without data leaving the network
Point LLM_API_BASE at a self-hosted model server, set STRIX_TELEMETRY=0 and leave remote tracing unset. The docs warn that most local models under 70B parameters struggle, so expect shallower results than with a frontier model.
Before you commit
- Setup
- Docker must be running. Install with the curl script, pipx, uv or pip (package strix-agent), set STRIX_LLM and LLM_API_KEY, then strix --target with a folder, repo, URL or API spec; the first run pulls the sandbox image.
- Integrations
- LiteLLM providers including OpenAI, Anthropic, Vertex AI, Bedrock, Azure, OpenRouter, Vercel AI Gateway and local servers; GitHub Actions and other CI; OpenAPI and Postman targets; your own MCP servers; nine agent skills via npx skills add usestrix/strix; optional Exa or Perplexity web search.
- Data and privacy
- Runs locally in a Docker container; results and traces stay in strix_runs/ unless you configure a remote OpenTelemetry endpoint. Code and target responses go to the model provider you choose. Telemetry (PostHog, Scarf, OpenTelemetry) is on by default. The local viewer binds to 127.0.0.1 with a tokened link. Strix Cloud shows SOC 2 and ISO 27001 badges; its retention terms are not published on the pricing page.
- Cost at scale
- Free software plus model usage per scan; no per-scan token counts are published, and deep scans cost the most. --max-budget caps estimated spend across all agents. Strix Cloud Pro is $29 per seat per month, pentests $100 to $300 each in credits, and Enterprise is custom.
- Limits
- Deep scans take 1 to 4 hours; cost estimates can under-count for providers that do not report usage; black-box runs can waste turns on routes that do not exist; frontier OpenAI, Anthropic and Gemini models need cyber access approval; results need human confirmation.
What’s good
- Tests live apps black-box as well as source code
- Apache-2.0, with any LiteLLM provider or a local model
- Quick-mode diff scans and a failing exit code fit CI
- Keeps findings it could reproduce, with remediation advice
What’s not
- Actively attacks targets; safe only on authorised, disposable systems
- Model costs vary widely and the budget cap is an estimate
- Telemetry on by default; code goes to your model provider
- Vendor benchmark is from v0.4.0 on known challenges
Best for
AppSec and platform teams with staging environments and written authorisation who want repeatable, agent-driven testing of their own web apps and APIs without per-test pentest fees.
Strix alternatives
Strix: common questions
- Is Strix worth it?
- Strix is a capable, permissively licensed way to let AI agents test your own apps, provided you treat it as an active attacker on a budget and aim it only at systems you are authorised to test. We rate it 4.2 out of 5.
- How much does Strix cost?
- The CLI is free under Apache-2.0; you pay the model provider per scan. Strix Cloud Pro is $29 per seat per month with 50 PR reviews per developer, then $1 each; pentests are extra at $100 to $300 each, and a Full Audit starts at $2,000. Enterprise is custom-priced.
- What are the downsides of Strix?
- Actively attacks targets; safe only on authorised, disposable systems; model costs vary widely and the budget cap is an estimate; telemetry on by default; code goes to your model provider; vendor benchmark is from v0.4.0 on known challenges.
- Who is Strix best for?
- AppSec and platform teams with staging environments and written authorisation who want repeatable, agent-driven testing of their own web apps and APIs without per-test pentest fees.
- How much does Strix cost for a team or in production?
- Free software plus model usage per scan; no per-scan token counts are published, and deep scans cost the most. --max-budget caps estimated spend across all agents. Strix Cloud Pro is $29 per seat per month, pentests $100 to $300 each in credits, and Enterprise is custom.
- Is Strix safe for company data?
- Runs locally in a Docker container; results and traces stay in strix_runs/ unless you configure a remote OpenTelemetry endpoint. Code and target responses go to the model provider you choose. Telemetry (PostHog, Scarf, OpenTelemetry) is on by default. The local viewer binds to 127.0.0.1 with a tokened link. Strix Cloud shows SOC 2 and ISO 27001 badges; its retention terms are not published on the pricing page.
- What are the best Strix alternatives?
- The closest alternatives we have reviewed are Shannon, Codex Security CLI and SDK, PyRIT and garak.
Answers checked Nov 14, 2026.
Make Strix? Show this rating on your site
