Compare

Pentr vs Strix

By David Cristóbal, founder of Pentr · Updated 2026-10-21

Strix is an open-source (Apache-2.0) AI penetration testing agent that you run yourself, with Docker and an API key for a language model of your choice. It tests source code, repositories, live web applications and APIs, can be given credentials for authenticated testing, and runs in CI on pull requests. Pentr is a hosted service. It tests domains, IP addresses, CIDR ranges and URLs that your organization has proven it owns, runs without credentials, and seals the evidence of each completed run in a SHA-256 hash chain signed with Ed25519, which anyone can check offline with the open-source pentr-verify. Strix fits developer-side testing of code and logged-in application flows; Pentr fits testing whose output a third party, such as an auditor, has to be able to verify.

Pentr wrote this page. Statements about Strix come from its README and documentation, linked under Sources, as they read on 2026-10-21. Statements about Pentr come from the Pentr documentation.

Side by side

	Pentr	Strix
Delivery	Hosted service, used through a web console and an API. There is no self-hosted edition.	Open-source command-line tool that runs locally with Docker. Strix also offers a hosted platform, Strix Cloud, and an Enterprise offer with VPC or self-hosted deployment.
Licence and price	Commercial. From €150 a month, billed monthly, or €120 a month billed annually (pricing).	Apache License 2.0. The open-source edition is free; you pay your language-model provider.
Language model	Operated by Pentr. Inference for the agent is provided by Nebius, in Europe (subprocessors).	Chosen by you, from hosted providers or a local model, through the `STRIX_LLM` setting.
What it tests	Domains, IP addresses, CIDR ranges and URLs registered to your organization.	Local code directories, GitHub repositories, live web applications, and APIs described by an OpenAPI file or a Postman collection.
Source code	Targets are hosts, networks and URLs; there is no code target.	Static and dynamic analysis (SAST and DAST) are listed among its tools.
Authenticated testing	No. There is no way to pass credentials to the agent, so every run is unauthenticated (scope).	Yes. Credentials are passed to the agent as an instruction for grey-box testing.
Authorization to test	Every target must be covered by an asset whose ownership proof holds at launch: a DNS TXT record or a well-known file for a domain, an owner's attestation for an IP address, range or URL (ownership verification).	The README states that it should be run only against systems you own or have explicit, written permission to test.
How a run is organised	One agent works through seven phases in a fixed order, from passive reconnaissance to post-exploitation and reporting (phases).	A graph of specialised agents for reconnaissance, exploitation and post-exploitation, running in parallel across targets.
Run limits	500 agent steps per run. The scheduler runs one engagement at a time. A queued or running engagement cannot be cancelled or paused (FAQ).	A per-agent turn limit (`--max-turns`, default 500) and an optional cap on model spend in US dollars (`--max-budget`).
Evidence of the run	Each command, HTTP exchange and screenshot is hashed with SHA-256 when captured. A completed run is sealed: its artifacts are ordered into a hash chain and signed with your organization's Ed25519 key. The sealed locker can be exported and checked offline, with no account, using `pentr-verify` (MIT licence).	Results are written to a `strix_runs/` directory and can be browsed in a local web viewer. Signing or hash-chaining of results is not described in the README or the documentation index.
Reports	A PDF report and SARIF and CSV exports. Each finding is mapped to the controls of one of six frameworks; a control id beside a finding is a pointer, not a verdict (frameworks).	Findings with CVSS scores and OWASP classification, and reports described as compliance-ready. Enterprise adds custom reports for SOC 2, ISO 27001 and PCI DSS.
Automation	A partner API, and schedules that launch an engagement on a cron expression (schedules).	A GitHub Actions workflow that tests pull requests; a non-interactive mode that exits non-zero when vulnerabilities are found.
Business logic	Not covered. The OWASP WSTG business-logic category is not among the checks the agent is directed to (what Pentr is).	Business-logic flaws (race conditions, payment manipulation, workflow bypass) are listed among the classes it tests for.

When to use which

Strix is the better fit when

you want to test source code or a repository before it is deployed;
the application has to be tested behind a login, with credentials you supply;
you want a check on every pull request in CI;
you want to choose the language model, run a local one, or keep the whole run on your own machines;
you want an open-source tool at no licence cost.

Pentr is the better fit when

the results go to an auditor or a customer who needs to check that the evidence has not changed since the run, without trusting a PDF;
the targets are internet-facing hosts, IP ranges and domains that you can prove you own;
you want findings mapped to SOC 2, ISO/IEC 27001, PCI DSS, NIST CSF, OWASP ASVS or ENS controls;
you would rather not run the agent, its containers and its model yourself.

The two can be used together: Strix in CI against code and logged-in flows, and Pentr on a schedule against the deployed, internet-facing estate when signed evidence is needed.

What Pentr does not do

These limits are set out in the Pentr documentation and matter for this comparison:

Runs are unauthenticated; anything behind a login is out of reach unless the agent gets in by attacking the login itself (scope).
One engagement runs at a time, and there is no cancel, pause or abort (FAQ).
The agent sets each finding's status (verified, false_positive, not_exploitable); nothing checks that status against the evidence (findings).
A valid signature shows that the evidence has not changed since it was signed; it does not show that the test was complete (what verification does not prove).

Sources

Strix README, github.com/usestrix/strix (commit a3c3123), checked 2026-10-21.
Strix LICENSE (Apache License 2.0) (commit a3c3123), checked 2026-10-21.
Strix documentation: CLI reference, checked 2026-10-21.
Strix documentation index, checked 2026-10-21.
pentr-verify, github.com/PentrAI/pentr-verify (MIT licence), checked 2026-10-21.
Pentr documentation: Scope, Ownership verification, Phases, Plans and limits, Verifying a locker.