Product

Detect. Protect. Investigate.

One dashboard for live detection, policy and incident review.

Live detection across all four checkpoints, with attacks blocked and critical events flagged for review.

Choose Off, Monitor or Block for each check.

Every event is logged with the agent, layer and rule, ready to review and triage.

AgentHacker dashboard overview showing attacks blocked and detections per checkpointAgentHacker policy page with Off, Monitor and Block controls for each checkAgentHacker investigations list of high severity events by agent, layer and rule

Walkthrough

See it work in two minutes.

Where Guard sits in the agent loop, a blocked request at each checkpoint, then the same events in the dashboard and in Slack.

Silent two-minute cut of a recorded demo, using an order-status agent.
01Where Guard fits02Normal work03Four boundaries04Overview05Investigate and triage06Tune, test, publish07Spike and audit report08Slack alerts

The full walkthrough

7 MIN · WITH NARRATION
The same demo end to end, with narration.

Protect

Off, Monitor or Block, check by check.

Choose Off, Monitor or Block for each check. Each section shows how many times it checked in the last 7 days.

User input

Messages people send to the agent. Stops jailbreaks and disguised instructions typed by a user.

Blocking
Hidden-instruction detection
Catches disguised instructions: invisible characters, look-alike letters, encoded text.
OffMonitorBlock
Checked 373 times in 7 days
Tool & document data

Content the agent reads from tools, files and web pages. Stops instructions hidden inside documents, emails or web pages.

Blocking
Scan tool and document content
Checks everything the agent reads before it acts on it.
OffMonitorBlock
AI check for hidden instructions
A model reads tool and document content for instructions aimed at the agent.
OffMonitorBlock
Link checking
Checks links and addresses against the hosts you allow.
OffMonitorBlock
Checked 2,234 times in 7 days

Also in the dashboard

Built for the people who run it.

01

Incident review

Open any flagged event, then confirm the block, mark a false alarm or escalate it.

02

Policy management

Draft a change, test it against saved cases, then publish with a version history that records what changed and why.

03

Analytics

Ask Spike questions about your live threat data, and generate an audit report that maps findings to OWASP.

04

Slack and access

Alerts post to Slack with Correctly blocked, False positive and Escalate buttons. Access is role-based: owners, admins and viewers.

Next

See what is running inside the loop.

Four checkpoints inspect what the agent reads, calls and says.

How it works →Join the beta →