Waiting for the draft
Ready for an agent to send
Sent by a support agent
Send
Routing rules, and a limit your team sets for when the agent may answer alone.
Open tickets newest first
Ticket
Subject
Intent
Priority
#48213
Charged twice for November
ReadingBilling, refund
UnsetHigh
#48212
How do I add a second admin?
How to
Low
#48211
Export to CSV times out
Bug
Normal
#48210
Change the card on file
Billing
Normal
#48209
Cancel at the end of the term
Cancel
High
#48208
Single sign-on setup help
How to
Normal
#48207
Invoice shows the wrong address
Billing
Low
A test set of past tickets that gates every change.
A weekly report of tickets answered, routed and handed over.
A runbook your support team owns.
Typical building blocks: Claude Agent SDK behind the helpdesk webhook; retrieval through the enterprise search MCP; scoped tools; a guardrail; tau-bench style evals.
Alternatives: OpenAI Agents SDK, LangGraph or Mastra; buy Sierra, Decagon or Fin when the volume is generic.
How the weeks run
Live in about 5 weeks
The work happens in your cloud, with the people who will own it.
Week 1Read a sample of past tickets, agree how sure the agent must be to answer alone, and build the test set.
Weeks 2 to 4Connect the agent to your helpdesk in draft mode and tune it on real tickets.
Week 5Let the agent answer on its own for the first queue when it is sure, and hand over the runbook.
After thatRunning on real traffic
Tickets answered or routed, with the reason.
Before we start
Admin access to the helpdesk and an export of past tickets.
A named support lead who decides when the agent may answer alone.
About 3 hours a week from that lead, and from two people who review drafts.
Fit
Is this the right first build?
It fits when
Simple tickets wait behind hard ones.
Tickets land with the wrong team.
A team on your side will own it after we leave.
It does not fit when
You want a hosted tool picked and resold to you.
You want a strategy deck in place of working software.
Nobody on your side can own it after we leave.
What’s included and what isn’t
Included
One helpdesk and one queue, in one language.
Up to 20 ticket types answered or routed.
A test set of up to 300 past tickets.
A weekly report of tickets answered, routed and handed over.
Not included
Voice and phone support.
Rewriting your help articles.
More queues and languages, which are added after the handover.
Fixed quoteOne price, agreed before work starts
About 5 weeks. You get a price range before the call and one fixed number after it. That number is the invoice.
If the work takes longer than we planned, we absorb the cost.
You pay 50% at the start and 50% at acceptance. Builds of 6 weeks or more pay 40% at the start, 30% at the midpoint and 30% at acceptance.
If a build starts within 60 days of your Audit results, the full Audit fee comes off its price.
We start three new projects a quarter, so your start date depends on when you book.
We pick the one that fits the systems you already run. You own the config, the tests and the runbooks, so you can swap any part later. Most often the Claude Agent SDK behind your helpdesk webhook; OpenAI Agents SDK, LangGraph or Mastra when they fit better. When your volume is generic, we tell you to buy Sierra, Decagon or Fin instead.
Will it answer customers on its own?
Only when it is as sure as the limit you set. Otherwise a person answers, and you can start in draft mode, where people approve every reply.
Which helpdesks does it work with?
Any helpdesk that can send a webhook and accept a reply through its API.
Where does it find answers?
In your help articles and past tickets, through the same permission-aware search as the rest of your company knowledge.
How do you test it?
Past tickets with known good outcomes form a test set. Every change must pass it before it reaches customers.
Support triage agent, running in your cloud in about 5 weeks.
Tickets answered or routed, with the reason. You own the config, the tests and the runbooks.