Private beta

You know what your AI costs. Not what it's doing.

Lisran breaks your spend down by task, so you can see which ones earn their keep and cut the ones that don't — then checks the whole thing against what you were actually billed.

Free to start · 10-min setup

Cost by task
last 30 days

87% tagged

TaskCostPer run
summarise-ticket$4,610$5.48

runs on opus — reroute

draft-reply$2,856$2.83

runs on opus — reroute

nightly-report$1,565$11.18

runs on opus — reroute

extract-entities$1,281$1.32

runs on gpt-4-turbo

classify-intent$21$0.032

runs on 4o-mini

Three of these cost more than the work needs. One costs three cents a call and should be left alone.


01/

Cost per task, not just per model.

Lisran reads the tags your application already attaches to each request, so spend lands against the thing it was for: summarise-ticket, nightly-report, classify-intent. No provider billing API carries that dimension — they group by model, project and day, which tells you what you bought and never what you bought it for.

02/

What each run actually costs.

A task costing $11 a call and a task costing a tenth of a cent are different businesses, and the monthly total hides which is which. Lisran divides by the calls, so you can see which features earn their keep at the volume you're running them.

03/

The ones worth rerouting, named.

Where a task runs on a model bigger than its work needs, Lisran says so and attaches the monthly saving. “Spend less on Opus” is not a plan. “Route summarise-ticket to Sonnet, save $2,900 a month” is something a team can do on Monday.

04/

The storage hiding in your token bill.

AI datacenter demand has pushed enterprise SSD prices up several times over, and the storage you pay for per token moved the same way. Keeping a prompt warm costs a tenth of re-sending it — but writing that cache costs more than not caching at all, so a cache nothing reads back loses money. Lisran measures which one you're doing.

05/

Numbers that match the invoice.

Most tools estimate cost from token counts and drift from what you were actually charged. Lisran calibrates every day against your provider's billed cost, so the task breakdown and the invoice agree.

Split byTaskModelTeamMemberCustomerProductProvider

30 days

of spend backfilled the moment you connect a key

10 minutes

to set one up, start to finish

$0

in infra changes — Lisran sits on top of your stack

06/

Rather have us walk you through it?

Create an account and connect a read-only usage key, or leave your work email and we'll walk you through your own numbers first — broken down by team, customer and task.

  • Cost per task, and per run, from your gateway tags
  • Which tasks to reroute, with the monthly saving attached
  • Calibrated daily against what you were actually billed
  • Team, member, customer and product attribution too
  • Quarter and year reporting, not just last 30 days
  • Full CSV export, nothing locked in

07/

Questions.

Lisran connects read-only to your provider usage and billing APIs. We read token counts, model names, and timestamps to compute cost — never your prompts or completions. Keys are encrypted at rest and you can revoke them anytime.

Find out which half of your AI spend is working.