Practical overview
When it fits
Use it when an AI system needs repeatable adversarial testing across model prompts, agent workflow changes or multimodal inputs, especially where a command-line and configuration-based evaluation flow is useful.
Who it is for
- Developers testing LLM applications
- Researchers evaluating model behavior
- Security teams conducting AI red-team assessments
Documented uses
- Probe LLMs with jailbreak and fuzzing inputs
- Test systems that accept text, image or audio inputs
- Run configurable security checks in a CI workflow
- Load custom prompt datasets for evaluation
- Connect scans to an HTTP-described LLM API
Requirements & limits
The project is publicly available on GitHub, is open source under the Apache-2.0 license, and documents Python package installation. It can be run locally or as a server process.
- The target must be expressible through the documented plain-text HTTP request specification.
- Target application credentials are needed for authenticated API requests.
- Custom local datasets must include a prompt column.
- The configuration exposes scan budget and failure-threshold controls for CI use.
- Multimodal testing is relevant when the target accepts text, image or audio inputs.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up an Agentic Security scan target
Install the tool and configure an HTTP request specification for an LLM security scan.
Before you start
- Python environment with pip
- An LLM endpoint address and application credentials
- A request body containing a <> placeholder
Steps
Install the package with pip install agentic_security.
Create the target specification using the documented HTTP method, URL, headers and request body format.
Start Agentic Security with agentic_security or python -m agentic_security.
Use agentic_security --help or set the documented host and port options to review or adjust startup settings.
Expected outcome
A locally started Agentic Security service is configured to send attack prompts to the specified LLM endpoint.
Guide limitations
- The documented integration template uses a plain-text HTTP specification; additional integration templates are marked as to be determined.
- The source describes startup and configuration, not a successful scan against a particular target.
Prepare custom prompt data for a scan
Add local prompt cases that Agentic Security can load during startup.
Before you start
- One or more CSV files
- A prompt column in each CSV file
- An Agentic Security installation
Steps
Create CSV files containing the custom cases.
Place the files where Agentic Security loads local CSV data.
Ensure each file includes a column named prompt.
Start Agentic Security and review the reported local CSV files.
Expected outcome
The tool detects eligible local CSV files and makes their prompt data available to the startup workflow.
Guide limitations
- Files without a prompt column are reported as invalid for this loader.
- The source does not document the result of executing a scan with custom data.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 3 October 2026
- Original website title
- msoedov/agentic_security · Official homepage title · retained in its source language