AI routing infrastructure · LLM Cost Intelligence

Your Stack.
Your data.
Any LLM.

Keep the tools you run, keep your data where it is, switch models whenever you want. Casca does three things: stops secrets before they are sent, leaves a trail you can explain, and sends every AI call to the model that actually fits.

What happens on one pipeline
AIShutter"Customer A123456789 is 30 days overdue" → sent as [TW_ID_1]real value stays local
GateEvery call: time, unit, rule, modelaudit trail
AI Route"Where is order #4471?" → simple lookupGemini Flash
AI Route"Assess the risk in this credit file" → needs reasoningClaude Sonnet
0original values leave your computer (AIShutter)
19countries and regions of local PII formats
<1 msclassification time per request
30–75%LLM bill reduction (by industry)
Three products, one engine

Each solves one thing.
They also stack.

The same multilingual classification and PII engine, packaged three ways. You do not have to use them together: each stands alone, and when stacked the rules and placeholders stay consistent. Pick the one you are curious about.

01 · AISHUTTER

Mask it
before you send it.

The AI privacy shutter on your computer. The moment you press Send it finds ID numbers, card numbers and secrets, shows you a confirmation card, and placeholders go out instead; replies are restored locally.

  • Free browser extension; the Windows build adds desktop AI apps and the CLI
  • No backend, no telemetry; real values never leave your computer
  • Enterprise: a management center for policy, groups, statistics and audit
Learn about AIShutter →
0original values leave your computer
02 · GATE

To get AI into the bank,
get it past the audit first.

A bank-grade LLM governance gateway. PII is masked at the gate, every call leaves an exportable audit trail, and cost discipline says: when unsure, choose the stronger model.

  • Multilingual PII detection and policy enforcement, tolerant of full-width and evasive forms
  • Decision trail exported as an audit evidence chain in one click
  • Self-hosted in your own data centre; data never leaves
Try the interactive demo →
Every callleaves an exportable audit trail
03 · AI ROUTE

Cut LLM bills 30–75%.
Change one line of code.

A research-backed routing engine. Every request is classified HIGH / MED / LOW in under 1 ms and sent to the model that fits, so simple questions stop paying GPT-4o prices.

  • 30–75% bill reduction by industry, public technical paper
  • Quality SLA in the contract: three days below threshold, full refund
  • OpenAI SDK compatible; change one base_url and you are live
Learn about AI Route →
30–75%bill reduction by industry, backed by a published paper

Vast Intelligence Limited

A Taiwan company based in Taipei, founded by Jeremy Chen. The technical paper (Zenodo DOI) and reference implementation are public on GitHub. Tell us which product you want to talk about.

casca@vastitw.com · +886 2 2706 7590