# Support Chat & Escalation

## What is it?

The support chat is the fastest way to get help: you ask your question, and
an AI assistant answers — usually right away, based on this help
documentation. If the assistant can't or isn't allowed to help itself, it
hands the request over to a human from our team. The entire history carries
over, so no one has to ask their question twice.

You'll find it as a **chat-bubble icon in the header** — top right, next to
search. Clicking it opens support as a panel on the right edge; you don't
have to leave the page you're working on to use it. If you'd rather browse
the docs yourself or look something up, the same content is searchable in
the [Help Center](https://www.testlabiq.com/en/help/help-center).

## How to use it

1. **Open support:** Click the chat-bubble icon in the header. You'll see
   your previous requests.
2. **Ask a question:** Click "New request" and describe your issue. The
   assistant searches the help docs and replies with the matching section.
3. **Follow up:** If the answer isn't enough, just keep writing in the same
   thread. The assistant knows the request's history, so "and how do I delete
   it again?" works without restating the whole question. For a *different*
   issue, open a new request — there it starts without the old history.
4. **Closing:** Once your issue is resolved, you can close the request with
   "Close".
5. **Request a human:** Write "talk to a human" — the request is immediately
   handed over to the team.

### What the assistant can access

So it can not only explain in general but also look things up concretely
("is my project X running?", "which station is free?", "why isn't my webhook
arriving?"), the assistant is allowed **read-only** access to your
organization:

- **Domain data:** Projects, test runs, stations, test objects,
  nonconformances, inventory, and calibration status.
- **Settings:** which features are enabled, which webhooks and API access are
  configured, and whether they last worked.
- **Audit log:** who last changed what — this answers most "it stopped
  working since yesterday" questions without needing to ask.

The following always applies:

- **Only your organization.** The assistant technically cannot see any other
  customer's data — the organization is fixed for the entire session and
  can't be changed by a question or a trick.
- **No secrets.** It never sees webhook signing secrets, API keys, or tokens
  — it only knows their names and identifiers. So it couldn't name such a
  secret even if asked to.
- **Its own role, always the same.** The assistant works with its own role.
  It doesn't depend on who's asking: as an admin, you see no more through the
  assistant than a colleague with viewer permissions — and no less either
  way. Beyond reading, it has a narrowly scoped ability to create things (see
  below) — it can't change or delete anything, and it never touches
  organization settings (webhooks, roles, feature flags).
- **Traceable.** Which queries it made is shown on the respective chat
  message and in the audit log (actor type *Agent*).

### What the assistant can create — always with your confirmation first

On request, the assistant can propose creating a **test station, a test
object, a piece of equipment, or a customer** — for test stations, test
objects, and equipment, including a matching **type**, if none exists yet.
For example: "create a new test station of type tensile tester" or "create a
new customer Acme Inc."

If you name the **property fields** you want when creating a new type (e.g.
"...of the new type tensile tester, with serial number and calibration
interval in months"), the assistant creates them as real, typed fields on the
type — not just as free text in the description. You can later add to or
change these fields as usual in Settings under "Types".

If you ask for **several tasks in one message** (e.g. "create the station
types A and B"), the assistant proposes a separate card for each task — you
confirm or decline each one individually.

Here's how that works:

- **Nothing happens automatically.** Each proposal appears as its own card in
  the chat with the concrete details (name, type, status, properties if any)
  and two buttons, **Confirm** and **Decline**. Only clicking "Confirm"
  actually writes anything — until then, nothing changes, and the other cards
  from the same message are unaffected.
- **A proposal expires once you keep writing.** If you reply about something
  else instead, all still-open cards from that request are treated as stale
  and can no longer be confirmed (they then show "No longer current") — the
  assistant will make fresh proposals if needed.
- **This is traceable too.** A confirmed action shows up in the audit log as
  actor type *Agent*, on behalf of you as the person who confirmed it.
- **Everything else the assistant cannot create, change, or delete** — for
  that, use the app as usual.

### When a human automatically takes over

The assistant escalates on its own when:

- you explicitly ask for a human;
- it concerns **contracts, billing, data deletion, or legal matters** — such
  topics are always decided by a human, never by the assistant alone;
- it finds **no matching basis** in the help docs (in that case it doesn't
  guess, it honestly hands over).

### What happens after the handover

Once a request goes to our team, **we reply by email** — to the address on
your account. Please reply directly to that email; further correspondence
happens there.

The handed-over request in the chat is then **closed**: you can no longer
write in it. That's intentional — a message there would no longer reach our
team, since the case now lives in our ticketing system. The history remains
visible for reference.

**You can open a new request at any time:** just start a new one. That's the
right way to raise a different topic — not by following up in the handed-over
request.

### The assistant sometimes reaches out on its own

If your organization's setup isn't finished yet, the assistant can proactively
open a request and offer to help with the next step. This happens at most
once — you can simply reply or close the request.

## Common issues

- **"The assistant doesn't understand my question."** Phrase it with
  concrete terms (e.g., "API key", "calibration", "webhook"). If that doesn't
  help, request a human.
- **"I can no longer write anything in my request."** Then it was handed over
  to our team. The reply comes by email — please reply to that. For a
  different topic, open a new request.
- **"No more replies are coming in the chat."** After the handover, we reply
  by email, not in the chat. Check the inbox of the address you're logged in
  with (including the spam folder).
- **"My request is closed, but I have another question."** A closed request
  no longer accepts new messages — just open a new one.
- **"I don't see a colleague's request."** Everyone only sees their own
  support requests. That's intentional: histories can contain personal
  information.
- **"The Confirm button on a proposal is gone, I only see 'No longer
  current'."** You kept writing after the proposal instead of confirming or
  declining it, so it's now considered stale. Just describe what you need
  again — the assistant will make a new proposal.
- **Transparency:** Every automatic action by the assistant is logged (actor
  type *Agent*, see the "audit-log" help page) and attributable to a human.

---

Source: https://www.testlabiq.com/en/help/support-chat