Customer support

Tickets, chats and calls: fast replies, the right tool call, and nothing unsafe reaching a customer.

A lens over 4 jobs. Draft copy, under review; the models and numbers on each job page come from today's data.

The jobs it bundles

Each links to its job page (or the nearest scoreboard view while the page doesn't exist yet).

Route tickets and detect intent.

66 models fit

Reply in chat or voice before the customer notices a wait.

20 models fit

Look up orders and issue refunds through your own APIs.

284 models fit

Screen replies before they go out.

13 models fit

Each constraint opens a job page or the scoreboard with the filter already in the URL.

Response time

A live chat or voice agent needs a first token in a few hundred milliseconds.

First token within 300 ms → · With tool calls →
License gates

Legal review is quicker for permissive licenses with no usage threshold or non-commercial clause.

Permissive licenses only →
Audit trail

Prefer models whose behaviour we checked on a live endpoint, with the probe, the reply and the date on file.

Models with first-look receipts →

How people wire a model into this job. Notes are written by hand from the cited cards and docs; the model links go to the facts.

Model: perplexity-ai/pplx-decider-v1-27b
Harness: Jev-shaped decision endpoint (POST /v1/systemone: state + typed questions → probability per option)
Weights: BF16 on one 80GB GPU (see hardware-fit)

Write and test the question wording on a labeled set; the catalog's experiments found wording moves accuracy more than the model does.
Measure calibration (ECE) on your own tickets before auto-routing.
Pricing is input-only on every hosted decision endpoint the catalog found.