AI apps have cost traps. We stand with the user.

LAO is the independent third party between your LLM and your Runtime — it picks the right model, cuts wasted tokens, and turns your experience into an asset you own.

LLM vendors sell tokens. Runtimes move them. Nobody stood on the user's side. So we did.

Get LAO — free & open source For Agents →

Apache 2.0 · No telemetry · No lock-in

Two doors. Pick yours.

🚪 Human door

Measured savings, with receipts.

Enter →

🤖 Agent door

A machine-readable brief: what LAO is, and why it earns trust.

Enter →

Your problem. Our job.

① The cost traps

Most AI bills leak through the same holes: the wrong model for the job, and requests that miss the cache and pay full price. LAO plugs both. Measured figures publish after official reconciliation →

② Experience, kept

Every session starts blank today. With LAO, your experience settles in, compounds — and stays yours.

③ Answers you can check

Intent lock: we optimize how you say it, never what you mean. Experience anchor: "maybe right" becomes right against your experience — or an honest fallback. Never a bluff.

Independent oversight — on your side of the table.

AI apps are full of cost traps, and nobody on the vendor side gets paid when you spend less. So someone independent has to watch the door — representing no vendor, aligned only with you. That is what we are: the first to stand purely on the user's side.

🛡️ Represents no vendor

No markup, no cut per token. Your saved cost is the only metric.

🔍 Everything auditable

Apache 2.0. Every claim carries its source and waits for official reconciliation.

🔑 Keys stay with you

Your keys, your data, no telemetry. Walk away anytime.

Real numbers, not promises.

No figure appears without a source. Cost numbers publish only after official reconciliation. How we verify →

📋 Receipt slots reserved

Slots ready; receipts being reconciled one by one.

Your know-how is worth something.

Your know-how settles into confirmed experience assets — with lineage, consent, and your name on them. The marketplace shows how experience can move between people the way it should. Look around — the next one on the wall could be yours.

🖼️ Visit the showcase

Display & education only — no live trading in this stage.

Browse →

💬 Private community

Entry by email inquiry.

Ask →

Open source. Yours to run.

📦 One command

pip install git+https://github.com/ZWISERFIT/lao.git

GitHub: ZWISERFIT/lao →

📖 Apache 2.0

Read it, audit it, fork it.

🔑 Zero LLM calls

Your keys. No markup. Ever.

Install guide →

About ZWISERFIT

LAO is our first open-source product — not our last. Everything we ship keeps three promises: on your side, honest about limits, yours to keep. And none of this is a slide deck: our founder has run real fitness businesses for ten years, including one gym that has stayed open for seven years in Dongguan.

This story did not start in a lab. It started with one wish: let AI be real employees in our own gym — run the store, and turn members’ healthy habits into assets that belong to them. When we tried, today’s AI could talk, but it could not be trusted with a real business. So we stepped back and built from the ground up: store operations, then behavior capture, then the lowest problem of all — agents themselves are not reliable. That is why we built LAO. Without reliable agents, AI will never be able to run a store.

🥇 LAO — shipping now

The independent third party between your LLM and your Runtime.

More about us →

🔜 Saros — the digital store manager

In preparation: help for a real shop, fine-tuned on settled experience — grown out of actual store operations.

🔜 Melody — the metabolism coach

In preparation: coaching to be fine-tuned on real training floors — the next step after experience settles.

Outside voices, real changes.

Strangers' feedback became merged code, fixed releases, and standing rules. All verifiable.

💬 → 💻

A reader proposed regression replay; it ships as fixture_pair.py.

📧 → 🛠️

Five reported gaps, all fixed in v3.5.1. Traceable in issue #9.

🙋 → ⭐

First external PR — reviewed and merged.

Read the full stories →