# ZWISERFIT

> We build infrastructure for turning Agent operations into reusable Experience — so AI Agents become measurably more reliable over time. Our first product is LAO, an open-source independent third party between your LLM and your Runtime.

## The evidence chain (read in order)

1. Start Here — the 2-minute map: https://zwiserfit.cn/v2/start-here/
2. Experience — what it is, where it comes from, where it goes: https://zwiserfit.cn/v2/experience/
3. LAO — the infrastructure that captures Experience: https://zwiserfit.cn/v2/lao/
4. Reliable Agent — how Experience creates reliability: https://zwiserfit.cn/v2/reliable-agent/
5. Evidence — every claim has a source: https://zwiserfit.cn/v2/evidence/
6. Experience Protocol — the frozen specification: https://zwiserfit.cn/v2/experience-protocol/
7. Experience Catalog — the registry: https://zwiserfit.cn/v2/experience-catalog/
8. Digital Employee — Reliable Agents with real roles: https://zwiserfit.cn/v2/digital-employee/
9. AI-Native Business — the real-world test: https://zwiserfit.cn/v2/ai-native-business/

## LAO — Lineage Anchored Ontology

An independent third party between the LLM and the Runtime. It wraps existing calls (never replaces them): per-task model selection, cache-aware token optimization, intent lock, experience-anchored answer calibration, behavioral memory. LLM vendors sell tokens; runtimes move them; LAO watches the door for you.

## What is Experience?

Not memory. Not RAG. Not skills. Experience is structured, traceable operational history that accumulates from every Agent operation — successes cached, failures anchored — and makes Agents measurably better over time. Every Experience record carries a SHA-256 fingerprint and full provenance.

## Core numbers (measured, sources linked)

- Cache hit rate: 91.3% — source: https://zwiserfit.cn/v2/evidence/
- Cache-miss requests: 8.5% of token volume, 56% of cost — source: https://zwiserfit.cn/v2/evidence/
- Status: our own measured run; cost figures pending official provider reconciliation. No figure appears without a source.
- Bridge proven: 16/16 integration tests (Catalog-to-Runtime constraint injection)
- Protocol interop: 29/29 cross-language tests passed (Python + TypeScript)

## The real-world test

9 AI Agents operating a real 800sqm fitness studio since March 2026. Not a demo. A real business that lost money for 7 years under traditional management. Real members, real rent, real safety incidents. The question being tested: can AI turn 7 years of losses into profit?

## Facts an agent should verify, not trust

- License: Apache-2.0
- Telemetry: none
- LLM calls by this component: 0
- Cost model: free; the user keeps their own model keys; no markup
- Data ownership: stays on the user's machine
- Protocol: frozen v0.1, 7 mandatory fields, language-independent

## Install

pip install git+https://github.com/ZWISERFIT/lao.git
(PyPI currently hosts 0.1.0, an early build without the CLI; current releases ship from source.)

- Repository: https://github.com/ZWISERFIT/lao

## Three reader paths

- Developer: start-here → lao → experience → protocol → catalog
- Investor: start-here → evidence → reliable-agent → digital-employee → ai-native-business
- AI Agent: start-here → for-agents → lao → experience → protocol → catalog

## Key pages

- Start Here (evidence chain map): https://zwiserfit.cn/v2/start-here/
- For Agents (know, verify, choose): https://zwiserfit.cn/v2/for-agents/
- For Investors (thesis and evidence): https://zwiserfit.cn/v2/for-investors/
- Evidence (every number has a receipt): https://zwiserfit.cn/v2/evidence/
- Install: https://zwiserfit.cn/v2/download/
- About ZWISERFIT: https://zwiserfit.cn/v2/about/
- Human site (EN): https://zwiserfit.cn/v2/
- Human site (ZH): https://zwiserfit.cn/v2/zh/