Skip to main content
Greyquill
Back
Product · Activate tier

GQ Agents.
Every result passes
through a verification harness.

An agent only sees what its playbook names. A threshold decides when a person steps in. And a pass is something an independent harness found by checking the actual result.

Book a walkthroughSee it in production
Playbook · claims-triage
scope
Claimall attributes
amount, status, filed_at, adjuster_id …
PolicyHolderfull_name, address
masked · tok_9f21ac30, never the value itself
Every other attribute on PolicyHolder: not handed to the agent, in any form.
An attribute not named here never reaches the agent, masked or not.
The dilemma it solves

Agents are easy to demo. Hard to defend.

When the regulator or the board asks what an agent did and why, most teams can only hand over a transcript.

95%

of GenAI pilots deliver zero P&L impact (MIT, 2025).

63%

of organisations have no formal AI governance in place (IBM/Reco, 2025).

What GQ Agents offers

Everything it takes to run an agent you can defend.

Six things happen around every action an agent takes. The three below have a visual, because they're the ones worth seeing rather than reading about.

01

Create

Define a playbook once: what it does, what it may read, what triggers it. Portable and declarative.

02

Compose

Agents invoke each other through packets. No hidden hierarchy decides who does what.

03

Scope & mask

An agent sees exactly the attributes its playbook names, and nothing else.

04

Gate

A threshold decides when a person has to sign off before the run continues.

05

Verify

An independent harness checks the actual result the run produced.

06

Observe

Token spend, cost, and latency, visible per run and per step as it happens.

How it works

The same playbook, two decisions made before it runs.

Playbook · claims-triage
scope
Claimall attributes
amount, status, filed_at, adjuster_id …
PolicyHolderfull_name, address
masked · tok_9f21ac30, never the value itself
Every other attribute on PolicyHolder: not handed to the agent, in any form.
An attribute not named here never reaches the agent, masked or not.
Same playbook · the gate
Claim.amount is ₹62,400 — above the ₹50,000 threshold
Run pauses · routed to gqdata:stewardA person decides
Under ₹50,000, the same gate lets the run continue on its own. The threshold lives in policy, so changing it doesn't touch code.
Independent verification

We don't take an agent's word for it.

Every check that decides whether a run passed is tested against a correct answer and a wrong one before it ever grades a real run. If a lazy or wrong result can slip through, the check doesn't ship. That discipline, built for our own agent workforce, is what a pass on your workload means too.

How a pass earns its name
1
Agent finishes a step
and reports done
2
An independent check runs
against the artifact it actually produced
3
Pass or fail is recorded
before the agent ever sees the answer
Before a check ships, a do-nothing agent has to fail it and a correct answer has to pass it, every time. A check that lets a lazy answer through never goes live.
Where it fits

The activation tier over a trusted foundation.

GQ Agents runs on top of GQData and asks GQ Govern's gates before every governed step. Without trusted data and an enforced policy, an audit-ready agent is still wrong, just provably wrong. GST Co-Pilot is what GQ Agents looks like in production for a real regulated workload.

GQData· FoundationGQ Govern· GovernGST Co-Pilot· Indian tax

Ship agents you can answer the regulator about.

Bring a workflow you wish you could automate but can't defend yet. We'll walk you through how GQ Agents would handle it, end to end.

Book a walkthroughSee the production proof