Ticket triage and routing
Reads each new ticket, tags its topic and sends it to the team that handles it.
Zendesk
Intercom
Customer support agent library
Agents for every part of support.
Each of these is a sub-agent of your Support agent: built from the work your team already does, approved by its owner, and run under the vault's controls. Browse by workflow, or search for the job you have in mind.
- Parent agent
- Support agent
- Example sub-agents
- 35
- Workflows
- 6
01 · Workflow
Triage
Every ticket in the right queue, with the right priority, in seconds.
Priority and SLA tagging
Sets priority from the customer's plan and the issue, and starts the right SLA clock.
Salesforce
Spam and auto-reply filtering
Closes spam, out-of-office and auto-replies so they don't sit in the queue.
Duplicate merging
Spots tickets about the same issue from the same customer and merges them.
At-risk flagging
Flags tickets from accounts that are renewing soon or already unhappy, and tells the account owner.
HubSpot
Slack
SLA breach warnings
Warns the team lead about tickets close to breaching their SLA.
Translation
Translates tickets in other languages for the agent and drafts the reply back in the customer's language.
02 · Workflow
Answers
Faster, consistent replies taken from what your team already knows.
Account and order lookups
Adds the customer's order, account and subscription details to the ticket before an agent opens it.
Stripe
Answer drafts
Drafts a reply from your help centre and past solved tickets, citing what it used.
Confluence
Repeat question replies
Answers simple, repeat questions, such as delivery times or how to reset a password, from approved replies.
Macro suggestions
Suggests new or updated macros from replies your team writes again and again.
Knowledge gap finder
Finds questions your help centre doesn't answer, from the tickets that needed a human.
Notion
Help-centre article drafts
Drafts a new article from a resolved ticket thread for the support lead to publish.
Stale article checks
Checks help articles against release notes and flags the ones that are now out of date.
03 · Workflow
Refunds & billing
Billing questions answered and refunds checked, with money moving only on approval.
Billing query replies
Answers questions about invoices, charges and plan changes from the billing system.
Refund checks
Checks a refund request against your policy and the customer's history, and recommends yes or no.
Credit note prep
Prepares the credit note for an approved refund or service credit.
Xero
Plan change requests
Handles upgrade, downgrade and cancellation requests: checks the terms and prepares the change.
Chargebee
Failed payment follow-up
Emails customers whose payment failed with a link to update their details, and tells the account owner.
Chargeback evidence pack
Gathers usage, emails and receipts into an evidence pack for a disputed payment.
Gmail
04 · Workflow
Escalations & incidents
When something breaks, everyone who needs to know knows, in words they understand.
Escalation summaries
Summarises a long ticket for engineering or a manager: what happened, what's been tried and what's needed.
Jira
Bug reports to engineering
Turns a confirmed bug into a well-formed issue with steps, examples and affected accounts.
Linear
Escalation chaser
Chases open escalations with engineering and tells the customer's agent when there's news.
Incident updates to customers
Drafts status updates for affected customers from the incident channel, on a set rhythm.
Statuspage
Customer incident reports
Drafts the customer-facing incident report from the internal post-mortem.
Complaint handling
Logs formal complaints, tracks the response deadline and drafts the reply for the complaints owner.
Google Docs
05 · Workflow
Quality
Knowing how support is doing without reading every ticket.
Weekly trend report
Reports volume, contact reasons, response times and what changed this week.
Contact reason tagging
Tags closed tickets with a consistent contact reason so trends are accurate.
CSAT review
Reads low-score surveys with their tickets and groups the reasons for the support lead.
Google Sheets
QA sampling
Scores a sample of each agent's tickets against your quality rubric for the team lead to review.
Coaching examples
Picks examples of great and missed replies for each agent's next 1:1.
06 · Workflow
Customer feedback
What customers ask for, gathered from every ticket and survey and handed to product with the accounts behind it.
Feature request log
Pulls feature requests out of tickets, de-duplicates them and counts who's asking.
Product feedback digest
Sends product a weekly digest of the top issues and requests, with the accounts behind each.
Survey follow-up
Groups survey comments into themes and routes unhappy customers to their account owner.
SurveyMonkey
Close-the-loop updates
Drafts a note to every customer who asked for a feature when it ships.
No sub-agents match . Discovery finds the jobs your team actually does, so yours will likely include ones not listed here.
Examples of what discovery typically turns into. Your library is built from the work your own team does, so names, triggers and connectors will differ.
How a library agent becomes yours
From the way your team works today to an agent in the vault.
01 · Discover
We find who does this work today.
Read-only access to your tools' audit logs shows where AI is already doing customer support work, and who is behind it. Nothing is installed on laptops.
02 · Interview
A short interview with the person.
They talk or type through how the job runs today. The output is a written spec: trigger, steps, tools, data and owner.
03 · Vault
Tested, approved, live in the vault.
The sub-agent is test-run on real data, its owner approves it, and it joins the Support agent under the vault's controls.
People ask the parent agent where they already work, in Slack, Teams, Claude or ChatGPT: @Support agent summarise the open escalations for Northwind
Controls
The controls support leaders ask about first.
Every sub-agent runs in the agent vault, with an activity log of every run, versions that need approval, and run history kept for 30 days.
- Money moves only on approval. Refunds, credits and goodwill gestures are recommended by an agent and approved by a person.
- Customer data on company terms. Tickets and customer records are read through company connectors in your region. Personal data in outputs can be tagged and held for review.
- Every reply has a trail. The activity log records each run: which tickets it read, what it drafted and which approved version ran.
FAQ
Questions from support teams.
Is this a fixed set of agents we switch on?+
No. These are examples of what discovery typically finds in support. Your library is built from the work your own team already does.
Will it reply to customers on its own?+
Only if you build it to. Most support sub-agents draft for a person to send, and external replies can be held for review per agent.
What if our help desk changes?+
Sub-agents use connectors, so the work is described independently of the tool. If a connector disconnects, the agent is marked down and its owners are asked to reconnect.
Keep reading
Related.
Book a demo
See the AI work already happening in your company.
Read-only access, nothing to install, first findings in days.