Agent fleet
Select a row for detail
| Name | Type | Status | Risk score | Tier | Actions |
|---|
Risk distribution
Live Threat Feed
streaming from the current scan
Blast Radius, Agent to Data Lineage
Which agents can reach which data sources based on declared tools
Risk by dimension
Fleet totals, in scorer points
Dimensions decompose the score emitted by
agentguard.risk.scorer.
Exposure is the no-authentication penalty (max 40).
Agency sums mutating, shell and tool-sprawl signals (max 55).
Data Quality is the orphaned/unknown status penalty (max 20).
OWASP covers credential exposure and read access (max 14).
The four sum to the raw score, which the scorer then caps at 100.
Asset × dimension heatmap
| Asset | Exposure | Agency | Data Quality | OWASP | Total |
|---|
All ten categories · AgentGuard detects LLM01, LLM06, LLM08
Categories follow the OWASP LLM Top 10 (2023 list), matching the tags emitted by the scorer.
Greyed rows are categories AgentGuard does not currently detect, absence of a finding is not evidence of absence.
🇪🇺 EU AI Act readiness
Applicable 2 August 2026
| Name | Owner | Risk category | Art. 9 | Art. 13 | Art. 50 | Disclosure | Status |
|---|
Derived from scan data, not a legal assessment.
Art. 9 (risk management) passes when the asset is below the high tier.
Art. 13 (information for deployers) passes when the asset's toolset is documented.
Art. 50 (transparency) is
not_applicable where no tool addresses a person,
and fails where a human-facing channel carries no declared disclosure.
Scanner configuration
GMS endpoint AgentGuard publishes to. Equivalent to
--url / $DATAHUB_URL.Stored in this browser's localStorage only. Never sent anywhere by this page.
One host:port per line. Mirrors the scanner's port list.
✓ Saved
These settings are held in this browser only, the page has no backend, so they do not reconfigure the CLI.
Pass
--url and --token to agentguard, or export DATAHUB_URL / DATAHUB_TOKEN.