The open-source AI builder platform you embed in your product.

Agents, apps, and skills, built on your servers with your LLM keys. Implement it yourself, or our AI-native engineers do it for you.

app.northwind.com/builder
N
PN

Arrears by building

Worst first · pulled live from the rent ledger · April fiscal year

Total arrears

$48,210

▲ $3,120 this month

Tenants overdue

31

9 over 60 days

Avg days late

23

▲ 4 vs last month

Outstanding by building

Harbour Court$14,760
Millbank House$10,915
Fenwick Rise$8,440
Alder Point$6,030
Castle Wharf$5,105
ChatVersions
v15 · idle
Show me arrears by building, worst first. Pull from our rent ledger.
Read schema ·rent_ledger
WroteArrearsChart.tsx
Installedrecharts

Done - arrears by building, worst first, live from the ledger.

Describe a change, or drop in a CSV or screenshot...
Claude Sonnet · company default

built with speculos-harness

Try the live demo

AI adoption was phase one. Owning the platform is phase two.

The mandate today is adoption - the questions that follow are cost, lock-in, and model choice.

The bill

Inference spend scales with adoption. Every generation here is metered - per person, per model, at provider rates.

Lock-in

Workflows built in a vendor's tool stay there. An open-source platform on your servers stays yours.

Model churn

The best model changes every few months. Here, switching is a configuration change - closed or open weights.

Open source, in three parts.

A React package, a Python backend, and a build service - working defaults out of the box, every piece swappable for your own. It runs on your servers, behind your SSO.

Frontendnpm

@speculos-harness/react

The chat, live preview, file view, and version history, dropped into your product as components.

npm install @speculos-harness/react
Backendpip

speculos_harness

The agent: a router you mount on your FastAPI or Flask service. Storage, auth, and models plug in here.

pip install speculos-harness
Build servicedocker

speculos/harness-bundler

A container that rebuilds the app on every change - fast enough that there's no run button.

docker pull speculos/harness-bundler

Instructions

set once by your admin
We are Northwind Property Group.
Amounts are USD. Fiscal year runs April-March.
Use the Northwind design system for every app.
Every table needs a CSV export.

Included on every build, company-wide.

Design system

your tokens, applied to every generated app

Models

a company default, plus what users may pick

SSO

your identity in front of the platform

Prompt log

every prompt recorded and replayable

Sandbox

generated code runs sealed; credentials stay on your server

spec_harness on GitHub

Your models, at provider prices.

No inference markup - usage is billed by your provider, on your keys, and metered per person and per model. Adding a model is a configuration change.

Spend·this month · by model
anthropic/claude-fable-5$1,412
openai/gpt-5.6-sol$486
moonshot/kimi-k3$94

budgets and rate limits: your LiteLLM proxy

Models·what users may pick
anthropic/claude-fable-5
openai/gpt-5.6-sol
moonshot/kimi-k3open weights
zai/glm-5.2open weights

Own the way your company builds with AI.

Our AI-native engineers set the platform up inside your infrastructure - and it stays yours to run.