Open-source agent
SwarmAttacker vs Strix
A VC-backed, Apache-2.0 agent graph that pentests web apps, APIs and code inside a Docker sandbox, with a paid hosted platform on top.
- Vendor
- OmniSecure, Inc.
- Licence
- Apache-2.0 CLI, proprietary hosted platform
- Pricing
- Free CLI; hosted pentests from $99 and audits from $2,000.
- Checked
- September 2, 2026
| Capability |  SwarmAttacker8 of 8 | Strix6 of 8 |
|---|---|---|
| Free | Yes. | |
| Runs on a ChatGPT subscription | Yes. | |
| Custom prompts & skills | Yes. | |
| Safety-refusal handling | Partial. | |
| Parallel multi-agent swarm | Yes. | |
| MIT-licensed open source | Not found. | |
| Black-box from a URL | Yes. | |
| Model-agnostic | Yes. |
- yes
- not found or only partly
Hover a mark for the evidence. A cross means the capability was not found, or only partly found, on Strix's public pages or repository on September 2, 2026, not that it is technically impossible.
Profile
What is Strix?
Strix is an Apache-2.0 autonomous penetration-testing agent from OmniSecure, Inc. in San Francisco, founded by Ahmed Allam and Alex Schapiro. The repository, created in August 2025, has about 60,000 stars; version 1.6.1 shipped on 2 September 2026 and the company reports a $6.7M seed round from June 2026.
The CLI runs agents inside a Docker sandbox with an HTTP proxy, browser automation, a terminal, a Python runtime, recon tools and code analysis. A root agent orchestrates and does no testing itself; it spawns specialist child agents that run in parallel, each loaded with up to five Markdown skills from the 76 that ship in the package. Targets can be URLs, GitHub repositories, local code, domains, IPs or OpenAPI and Postman specs. Runs are steered with an instruction flag or file, and MCP servers can be attached.
Models go through LiteLLM, so more than 100 providers work; the default is GLM-5.3 via OpenRouter, and a ChatGPT Plus or Pro subscription can be used through a Codex-style OAuth login that the code itself notes is not officially supported by OpenAI. The system prompt carries a refusal-avoidance section, but a provider refusal or ChatGPT guardrail block ends the agent with a suggestion to switch models.
The hosted platform adds scheduling, pull-request reviews, auto-fix PRs and integrations. Third-party reviews praise the breadth and criticise token cost and weak endpoint discovery on one real-world site.
In SwarmAttacker's favour
Where it stands out.
- 01MIT licence with no open-core split: everything SwarmAttacker does is in the public repository, while Strix keeps scheduling, auto-fix and billing in a proprietary hosted product.
- 02Refusals are handled, not fatal. SwarmAttacker detects a refusal and works through a retry, rephrase and model-fallback ladder; Strix ends the agent on a provider refusal or guardrail block and asks you to change models.
- 03No sandbox to build: SwarmAttacker runs with tmux and the usual CLI tools on your machine, so a first run is a clone, a `codex` login and one command.
- 04A fixed roster of OWASP and vulnerability-class specialists is dispatched by the planner, which keeps coverage predictable across runs instead of depending on what the root agent decides to spawn.
In Strix's favour
Where it is stronger.
- 01Broader scope: Strix also reviews repositories and local code, takes OpenAPI and Postman specs, scans pull-request diffs in CI and opens auto-fix PRs.
- 02A Docker sandbox with an HTTP proxy, browser automation and recon tooling isolates every run.
- 03A hosted platform with scheduling, history and integrations for teams that do not want to operate a CLI.
- 04Far more momentum: roughly 60,000 stars, near-daily releases and a $6.7M seed round.
Verdict
Which one should you pick?
Pick Strix if you want one tool that covers source code, APIs and CI pull requests, or you need a hosted product with a team workflow. Pick SwarmAttacker if you want a permissively licensed black-box agent that runs on the ChatGPT plan you already pay for, and keeps going when the model refuses.
Sources
What this page was checked against.
All checked on September 2, 2026.
Turn a URL into a security report.
Open source, MIT licensed, and it runs on the ChatGPT plan you already pay for. Install it and point it at a target you own.
brew install joloooo/swarm/swarm