Rogue
github.com
- Free plan
- Self-hosted
What Rogue is
Rogue is a self-hosted platform for evaluating AI agents and testing their security. Automatic checks compare agent behavior with business policies and expected behavior, then produce pass/fail reports with reasoning. Its red-team coverage spans 75+ vulnerabilities across 12 security categories, with techniques including encoding, social engineering, and injection; vulnerability scores use a CVSS-based scale from 0 to 10. Rogue maps testing to frameworks including OWASP LLM Top 10, MITRE ATLAS, NIST AI RMF, ISO/IEC 42001, the EU AI Act, GDPR, and OWASP API Top 10. It provides a server, terminal interface, and non-interactive CLI for CI/CD pipelines. Supported agent connections include A2A over HTTP, MCP over SSE or Streamable HTTP, and direct Python calls. OpenAI, Anthropic, and Google models are available through LiteLLM, and an LLM API key is required to start. Reports export to Markdown, CSV, or JSON, and scans can use a random seed for reproducibility. Basic scans use five curated vulnerabilities and six attacks; full scans use 75+ vulnerabilities and 40+ attacks.
Who it suits
Rogue is aimed at security teams and developers building or deploying AI agents, including those doing regression testing, security audits, or compliance reporting. It is self-hosted and requires an LLM API key to get started.
Good to know, and what to watch
- Produces pass/fail evaluations with reasoning
- Maps tests to named security and compliance frameworks
- Provides a CLI for CI/CD pipelines
- Exports reports in Markdown, CSV, and JSON
- Supports reproducible scans with a random seed
- An LLM API key is required to start
- Free use is limited to personal and internal use
- Commercial hosting requires licensing
Plans
Every plan and priceCommercial hosting
Personal and internal use
Free plan · pricing on request
Getting started
- Open the Rogue website or repository
- Install the prerequisites, including uvx and Python 3.10 or newer
- Provide an API key from OpenAI, Anthropic or Google
- Deploy Rogue in your self-hosted environment
- Choose the server, terminal interface or CI/CD CLI
- Select a basic or full scan and export the resulting report
Limits to check
Basic scans use five curated vulnerabilities and six attacks, while full scans use more than 75 vulnerabilities and 40 or more attacks. The free plan is for personal and internal use; commercial hosting requires licensing.
The verdict
Rogue combines agent evaluation, red teaming, and compliance-oriented reporting in a self-hosted tool. Its free terms cover personal and internal use; commercial hosting requires licensing.
Questions people ask
Is Rogue free?
The personal and internal use plan costs nothing. Commercial hosting requires licensing and contact with [email protected].
Which models does Rogue support?
Rogue supports OpenAI, Anthropic and Google models through LiteLLM. An LLM API key is required to get started.
Which agent protocols are supported?
It supports A2A over HTTP, MCP over SSE or Streamable HTTP, and direct Python function calls.
How can reports be exported?
Reports can be exported in Markdown, CSV and JSON formats.
Which interfaces are available?
Rogue provides a server, a terminal interface and a non-interactive CLI for CI/CD pipelines.
Who licenses Rogue?
The license identifies Qualifire ltd as Rogue's licensor.
What the maker publishes
- Commercial use
- The README says Rogue is free for personal and internal use and that commercial hosting requires licensing; it provides [email protected] for contact. source
- Compliance
- Rogue maps testing to eight compliance frameworks including OWASP, MITRE, NIST, GDPR and the EU AI Act. source
- Coverage
- The repository states that Rogue covers 75+ vulnerabilities across 12 security categories and 20 attack techniques. source
- Evaluation
- Automatic evaluation checks agents against business policies and expected behavior and produces pass/fail reports with reasoning. source
- Evaluation
- Automatic Evaluation tests agents against business policies and expected behaviors with pass/fail reports and reasoning. source
- Frameworks
- The documented framework coverage includes OWASP LLM Top 10, MITRE ATLAS, NIST AI RMF, ISO/IEC 42001, EU AI Act, GDPR, and OWASP API Top 10. source
- Interfaces
- Rogue provides a server, a terminal interface, and a non-interactive CLI for CI/CD pipelines. source
- Interfaces
- Rogue provides a server, terminal user interface and non-interactive CLI for CI/CD pipelines. source
- License
- The repository license is the Qualifire OSS License, combining MIT terms with a Commons Clause that excludes selling the software or offering it as a paid hosted service. source
- Maker
- The license identifies Qualifire ltd as Rogue's licensor. source
- Model providers
- Rogue lists OpenAI, Anthropic, and Google models via LiteLLM and requires an LLM API key to get started. source
- Models
- Rogue supports OpenAI, Anthropic and Google models through LiteLLM. source
- Product
- Rogue is an AI agent evaluation and red teaming platform for testing agent reliability and security. source
- Protocols
- The repository lists A2A over HTTP, MCP over SSE or Streamable HTTP, and direct Python function calls as supported agent protocols. source
- Protocols
- Supported agent protocols are A2A over HTTP, MCP over SSE or streamable HTTP, and direct Python function calls. source
- Purpose
- Rogue is an AI agent evaluator and red team platform for stress-testing agents before attackers do. source
- Red teaming
- The repository describes 75+ vulnerabilities across 12 security categories, with attack techniques including encoding, social engineering, and injection. source
- Red teaming
- Red Teaming simulates adversarial attacks to find security vulnerabilities. source
- Reports
- Rogue can export reports in Markdown, CSV, and JSON formats. source
- Reports
- Rogue exports comprehensive reports in Markdown, CSV and JSON formats. source
- Reproducibility
- Scans can use a random seed to make results reproducible. source
- Requirements
- The quick start lists uvx, Python 3.10+, and an API key from OpenAI, Anthropic, or Google as prerequisites. source
- Risk scoring
- Rogue assigns vulnerabilities a CVSS-based risk score from 0 to 10. source
- Runtime integrations
- The AI AppSec product page lists LangChain, CrewAI, AutoGen, custom builds, SIEM systems and webhooks, including Splunk, Datadog and PagerDuty. source
Checked against the maker's pages on 1 Oct 2026.