Live agent demo
Seven agents running a production pipeline: signed triggers, scoped identity, PII redaction, policy checks and signed human approvals. Pick one, send a request, and make the call when it asks you.
Order #4821 hasn't arrived · Sarah Khan, Email to support
09:41Payload
Sarah Khan <sarah.khan@outlook.com> My order #4821 was due yesterday and I need it by Friday. If it won't make it, I'd like a refund. Call me on 07700 900123 if easier.
- person → pending–
- email → pending–
- phone → pending–
Access token
No token until the trigger's signature checks out.
Trace
Spans appear here as each step runs, with timings from the OpenTelemetry trace.
Ready
Sample data in a scripted demo. Your agent runs on your own tools, identity provider, rules and approvers.
Support agent
Answer, resolve and escalate inbound requests across chat, email and voice.
Connects to
Stops for a person when
A refund over £50 pauses and goes to a named owner before it is issued.
How we measure the support agent
Every number comes from your own systems and the agent's audit trail, measured against a baseline taken before launch.
1Baseline
Before launch we measure the same numbers for two weeks, from your own systems, so results are compared, not claimed.
2Log every run
Each run writes to the audit trail: time taken, actions, tools touched, handovers and every human decision.
3Weekly report
A short report compares each metric with the baseline and lists the cases a person had to fix.
4Monthly review
With the workflow owner we tune thresholds, fix exceptions and decide what the agent takes on next.
| Metric | How we measure it | Data comes from |
|---|---|---|
| First response time | Median time from a customer's message to the first reply, before and after launch. | Helpdesk timestamps |
| Share of contacts resolved without a person | Tickets closed by the agent and not reopened within 7 days, divided by all tickets. | Audit trail and helpdesk status |
| Time to resolution on escalated cases | Time from handover to close, against the same case types before launch. | Helpdesk |
| Hours returned to the support team each week | Agent-resolved tickets multiplied by your pre-launch average handling time. | Baseline sample and audit trail |
| Customer satisfaction on agent-handled conversations | CSAT scores split between agent-handled and person-handled conversations. | Your CSAT survey |
Nothing runs unauthenticated
Signed triggers, short-lived tokens bound to the agent, and a scope for every tool.
Personal data never reaches the model
Names, emails, phones and bank details become vault tokens, and are re-identified only at delivery.
Rules before actions
A policy engine checks each plan, and an output guard checks every draft before it leaves.
People sign the big calls
Money, public messaging and unusual cases stop for a named approver who signs with SSO and a passkey.
Want one of these agents on your tools?
Book a free call and we'll map which agent fits your workflow first, and what it should stop and ask about.
We reply within one business day.