The same AI workload, before and after routing + caching
Twelve customer reviews, one sentiment label each, run twice over. The baseline sends every review to the connection's big model (gpt-5). The optimized copy says what kind of job it is (taskClass: sentiment), lets the routing profile pick the cheapest model that can do it (gpt-5-nano), and turns on the response cache, so asking the same question again costs nothing. The usage ledger shows both bills side by side. The model is a stand-in in the demo stack, so there is no key and no bill; the routing, pricing and caching are the platform's own.
The pipeline
This is the actual graph the template creates — 2 steps.
Connects to
- PostgreSQL
- AI model
Also includes
- 2 pipelines
How you know it worked
Run both pipelines once, then run demo-ai-cost-routing-cache-optimized a second time. Each run labels the same twelve reviews: 1, 4, 6, 8 and 11 positive; 3 and 10 neutral; the rest negative.
Try this templateUsed in: Retail & e-commerce
Related templates
Webhook → database
A pipeline that starts when something posts to it. The webhook's JSON body is the run's starting payload, so the first node already has the data —…
Sequence of pipelinesThree pipelines run as one unit (sequence)
A sequence is orchestration above the run machinery: it runs pipelines you already have, in order, feeding each one's output to the next. This one…
Published appOne app, one URL namespace (published app)
A published app groups things you have already published under a single URL — /apps/demo-shop/v1/… — with its own OpenAPI document. This one binds…
Stop moving data by hand.
Start free with three projects, or talk to us about running it across your organisation.