Score over time · every point is a real engine verdict
By domain · average score
Needs attention · flagged for human review sample data
Click any row to open the case review drawer.
Review status, assignee and SLA are demo workflow state held in this browser session. Scores, verdicts and report links are live engine output.
Every evaluation writes a permanent, shareable audit report on the live engine. Each report carries an immutable audit ID assigned at evaluation time; report content is never edited after issue.
Policy register demo edits stay in this session
Escalation thresholds · deterministic, computed in code
| Condition | Action |
|---|---|
| Any critical safety dimension at −7 or below | Critical flag + mandatory human review |
| Composite score below −30 | Human review required |
| Composite −60 or below | Reject and flag |
| Truth dimension at −5 or below | Hallucination: High flag |
Thresholds are computed identically for every evaluation. The critical safety dimensions are Restraint, Truth and Consequences; hover the term above for the underlying framework names. In the full product these policies are configurable per workspace.
Policy change history sample entries
How EthosGuard sits in your delivery path
communication
via live engine
approve · caution · block
or deliveryflagged cases routed
to a reviewer
with immutable ID
Every path ends in an audit record — including approvals that never touched a human.
Three integration patterns
Your systems call the evaluate endpoint before any AI-generated message reaches a customer. One POST per message; verdict, recommendation and audit link come back in the response.
EthosGuard sits around an existing copilot (service desk, CRM assistant, advisor chatbot) without changing it — responses are intercepted, scored, and released or held per policy.
Sample production traffic across many client AI systems into one governance view — posture metrics, drift and incident trails per business line.
The platform is API-first and structured for white-label deployment. Patterns beyond the embedded API are delivered per pilot engagement.
Connectors
Direct HTTPS evaluate endpoint. Live today — the API Access tab documents it.
Push alerts to your systems when a score crosses a threshold or review is required.
Salesforce, Zendesk, ServiceNow: open a review ticket automatically for blocked responses.
Stream evaluation records to Snowflake, BigQuery or Splunk for retention and analytics.
API status
● Operational — live evaluate endpoint at app.ethosguardai.com. Typical evaluation latency: 3–10 s for text (tiered routing), longer when escalation to the grounded engine occurs.
Every evaluation is assigned an audit ID at completion. Use it as your request reference: it is permanent, appears in the JSON response, and resolves to the public report URL.
API key demo
This is a masked placeholder — no real credential is stored in or served by this console. Per-client keys with usage metering ship with the pilot API tier; Reveal and Rotate here are demonstrations of that workflow.
Evaluate an AI response
Substitute your issued key for $ETHOSGUARD_API_KEY. The endpoint itself is live today; the Authorization header shows the pilot authentication pattern.
Response example
audit_id is permanent and public; request_id is your correlation handle for support and log matching. Recommendation values map to: Safe to act · Verify key claims · Use with caution · Reject and flag.
Webhook · score threshold alert roadmap preview
Threshold webhooks are on the pilot roadmap; the test button demonstrates the delivery feedback loop.
Documentation
· Methodology and trust: how the ten-dimension framework works and why the scores hold up
· Industry case library: thirty worked examples with verdicts and corrections
· API reference: the evaluate endpoint, report pages, and history
· The live engine: app.ethosguardai.com