Before: A response promises a refund or feature outside the customer's entitlement.
Policy travels with the answer
The assistant sees tier and entitlement limits before offering a remedy.
HINT for Customer Support
Your assistant replies fast. HINT keeps every answer inside support policy.
HINT keeps approved answers, entitlement rules, verification steps and escalation paths in plain-text notes beside support playbooks. Your assistant reads the rules for this product and customer context before replying.
The package is named for the role: @openhint/hintbook-support-agent.
Where it helps
Three situations your team will recognize
Before: Sensitive account details are disclosed before identity is checked.
Verification comes first
Required verification steps precede any action or information covered by the policy.
Before: A difficult case receives a confident invented ETA.
Unknown is an acceptable answer
The assistant states what is known and follows the declared escalation path without making a timing promise.
See it
A real support answer from the demo
The Vandelay Support demo grounds a LaTeX-product answer in an approved source.
demo-vandelay-support/products/latex/playbook.md.hint ↗
# read Approved source
../../sources/approved.md
# answer latex-1 {#support_agent_demo}
Approved answer grounded in SRC-1.
# never Invented detailsWhat your assistant does next
It replies from the approved playbook, verifies required facts, and escalates when policy does not supply an answer.
What your AI assistant receives
<answer name="latex-1">
Approved answer grounded in SRC-1.
</answer>
<never name="Invented details">…</never>Your boundaries stay visible
What it will never do
- Promise a remedy, feature or ETA outside declared policy.
- Disclose protected information before required verification.
- Invent a troubleshooting result or silently skip escalation.
FAQ
Questions support agents ask
Does it integrate with a helpdesk?
HINT governs the files and context behind an answer; your existing tools handle tickets and delivery.
Can it work in live chat?
Yes, when the assistant handling the chat can run HINT or receive its context through MCP.
Can policies vary by locale or tier?
Yes. Put narrower rules in the relevant product, locale or tier folder so they apply only there.
For your technical colleague
The vocabulary
A hintbook is a vocabulary for your profession—installed, not written by you.
product- Sets the product whose support rules apply.
tier- Defines the customer's support level.
intent- Names the customer's confirmed reason for contact.
answer- Records approved response content and its conditions.
entitlement- Defines what the customer's tier permits.
refund- Defines an authorized refund condition and limit.
identify- Sets required identity checks before disclosure.
escalate- Defines when and where the case must be handed off.
Manual setup
Bootstrap is read-only: it prints instructions for the assistant. The assistant performs the installation.
npm install -g @openhint/cli
hint config
hint add @openhint/hintbook-support-agent
hint apply
hint verify products/latex/playbook.md
hint emit products/latex/playbook.md