Practical overview
When it fits
Use HackAgent when you need structured, repeatable attacks against an AI model or agent and want judge-scored results to track exposure or compare outcomes after a fix.
Who it is for
- Developers testing the security of AI models or agents they own or are authorized to assess.
- Teams that want repeatable red-team campaigns and measured attack outcomes.
- Engineers integrating AI security testing into scripts or CI workflows.
Documented uses
- Define attack goals, a target, attack techniques and evaluation judges in a YAML campaign.
- Run a campaign and measure attack success using judge-model scores.
- Validate campaign configuration before running attacks.
- Browse campaign runs and results or automate workflows through the documented interfaces.
Requirements & limits
The documented API requires an API key created in the dashboard; some account endpoints require a signed-in dashboard session.
- Campaigns are YAML files that define the target, goals, attacks, judges and execution behavior.
- The same campaign format is documented for command-line, Python and terminal-interface use.
- Judge-model scores provide measured attack success rates; results depend on the configured judges and scoring method.
- The documentation identifies a hosted dashboard at app.hackagent.dev and an HTTP API for automation.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up and run a YAML campaign
Define a repeatable attack campaign and run it from the command line.
Before you start
- Python package installation with pip
- A YAML campaign specifying goals, target connection, attacks and evaluation judges
- A target model connection and judge-model connection
Steps
Install HackAgent with the command below.
pip install hackagentCreate a campaign YAML file with a version, campaign name, inline goals, target, attack and evaluation judge.
Run
hackagent campaign validate campaign.yamlto resolve and validate the campaign without running attacks.Run
hackagent campaign run campaign.yaml.
Expected outcome
The campaign runs against the configured target and evaluates replies with the selected judge scoring.
Guide limitations
- The documented example uses Ollama connections; the guide does not establish compatibility with other providers.
- Test only systems you own or have explicit permission to assess.
Install the web interface for browsing campaign results
Install and launch HackAgent's documented web interface.
Before you start
- Python package installation with pip
Steps
Install the web extra with the command below.
pip install 'hackagent[web]'Launch the interface with
hackagent web.
Expected outcome
The documented web interface is launched for browsing runs and results.
Guide limitations
- The documentation does not specify deployment or hosting steps for a self-managed web interface.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 7 November 2026