Use case

An AI agent for customer support that knows when to stop

The failure mode of support AI isn't a wrong tone — it's confident answers about things it was never told. Orqana agents answer from knowledge you attach and escalate everything else.

  • Every run logged
  • Revoke access instantly
  • Cancel any time
Cited answers

Grounded in knowledge you attach

Escalate first

Unknowns go to a human, not a guess

Per-agent scope

Support access isn't sales access

Grounded answers, not general knowledge

You attach the docs, policies and macros the agent may use. When the answer isn't in there, the agent says so and hands off rather than improvising.

  • Answers cite the source they came from
  • Unknowns escalate with a summary instead of guessing
  • Update the knowledge and behaviour changes immediately

The support jobs teams automate first

Start where volume is high and risk is low, then expand.

  • Repeat how-to questions and setup walkthroughs
  • Status and account lookups once the system is connected
  • Thread and ticket summaries for handoff between shifts
  • Draft-only replies for anything commercial or contractual

What stays human

Refunds, cancellations, security reports, legal threats, and anything involving a named account exception. Those are escalation rules in the brief, and they take precedence over everything else the agent was asked to do.

Start with draft-only support

Let the agent write every first reply and send none of them. Trust arrives faster than you'd expect.

Build a support agent

Audit and accountability

Support is one of the few functions where you may need to reconstruct exactly what a customer was told. Every agent run stores the input, the decision, the tools used and the output, and access can be revoked instantly per app or per agent.

  • Per-run history with timestamps
  • Per-agent tool scopes
  • Instant revocation from the integrations page

Cost shape

Support automation is high-volume, so run allowances matter more than seat counts. Starter is free to prove the pattern, Pulse covers a small team's steady volume, and Pro is sized for a support queue running all day. Pricing lists the exact monthly run allowances.

Common questions

Will the agent invent answers?

It answers from the knowledge you attach and cites the source. When the answer isn't there, the brief tells it to escalate with a summary rather than improvise — and you can see every one of those cases in the activity log.

Can it look up a customer's account?

Once the relevant system is connected, yes, within the scopes you grant that agent. Google Workspace is live today; helpdesk and CRM integrations are in build.

How do we keep it out of refunds and contracts?

Those become explicit escalation rules in the brief, and any write or outbound action can require approval. The agent's own judgement never overrides an escalation rule.

What does it cost at support volume?

Cost tracks monthly agent runs rather than seats. Starter is free for proving the pattern, Pulse suits a small team's steady volume and Pro is sized for an all-day queue; exact allowances are on the pricing page.

Describe your first agent today

Free to start. No credit card, no setup calls, no engineering ticket.

Get started free