FOR SUPPORT, OPS AND ENGINEERING TEAMS
Shrink the AI bill. People included.
Find which LLM steps can move to decision models or code. Compare machine spend, latency and the work that still reaches people — before you refactor.
MACHINES
PEOPLE
REACH A PERSON
p95 6.60 s
WHERE THE BILL GOES
█ machines 0.1% ▒ people 99.9%
C:\VARIA>RUN /ITEMS:10000 /TOUCH:3
01/05 THE FLOW AS DESIGNED. Triage asks three questions in one request. Refund requests go to an LLM that drafts a reply, and a second check catches over-promises.
How it works
C:\VARIA>TYPE HOWTO.TXT
- 1. COMPOSE
- The steps you would ship: decision-model questions, LLM steps, code rules, confidence gates, people.
- 2. SIMULATE
- 10,000 items with a fixed seed: cost per 1M, p50/p95 wait, the share that reaches a person.
- 3. WHAT-IF
- Change a share, a model or a rule. It runs again in milliseconds.
- 4. HAND OFF
- Request JSON for your repo, or let a coding agent edit the flow over MCP and the CLI.
What is VARIA?
VARIA is a decision-flow simulator for AI pipelines. You compose a flow of AI decisions: decision-model questions (Noul, Choice, Score), LLM steps, code rules, confidence gates and human handoffs. VARIA runs 10,000 items through it and shows cost per 1M items, p50 and p95 wait, decision calls per item and the share of work that reaches a person, before you build the pipeline.
Send your flow. Get the numbers back.
C:\VARIA>EXIT
VARIA is an independent project, not affiliated with TypeSafe. Jev is a trademark of its owner. Updated 28 October 2026.
IBM VGA 8x16 font: The Ultimate Oldschool PC Font Pack by VileR (int10h.org), CC BY-SA 4.0.