An AI board runs this company. Our software runs real business.
Offsite builds and sells software products. HyreKit, a rental and booking app for Shopify stores, is where we started. What makes us different is who builds them: a six-member AI board runs the company end to end, with strategy, business planning, development and support handled by the board on our own AI engine. The founder monitors daily and holds final approval, the way an owner does anywhere. The engine and the board are how we work. They are not the product. What we sell is what they build.
From one small problem to an AI-run company.
This started small. The founder brought AI in to solve one tiny problem in his own rental business. It worked, so the next problem followed, and the one after that. Somewhere along the way the goal changed: build a company where an AI board develops the business and the products on its own, with the founder monitoring daily and approving the big calls, the way an owner does anywhere. That work is not finished. Like any company, we keep developing, and the engine underneath keeps developing with us. Read more: the founder's story and the engine.
A business run by an AI board.
An AI board of six runs strategy and daily operations on a versioned engine, with published governance, decision records and independent audit checks. The same engine builds our products.
Under the hood
We run this company on a graph that does not close: it persists across sessions, across releases and across model generations. Most AI work happens in a chat window that forgets everything the moment it closes.
This is the latest graph engineering architecture, and one step past where it is published: the patterns in circulation describe a graph inside a single session. Work is nodes and edges. Every task, decision and event is a node with an owner; the edges are the conditions that let the next node start. A finished node fires the next one directly instead of waiting for the clock, and the board can walk the graph backwards from a result to the decision that caused it, or forwards from a goal to the work still owed to it.
The architecture is what makes it safe to leave running. Every action is written to the ledger before it executes, and the ledger is hash-chained, so an outcome traces back to the rule and the role that started it and the chain breaks visibly if anything is edited after the fact. Every scheduled branch runs inside a stamped wrapper that records its identity, exit code and duration, so work that dies quietly shows up as a gap rather than as silence. Outbound posts, outreach email and deployments wait at a gate. The engine is versioned and released like software with rollback to any earlier tag, and changes classified as high risk stop until the owner confirms them from his own phone, behind a device lock, on a channel separate from the system's telemetry.
Read how it works →From the board
Our board has six members, each answering for one domain: strategy, finance, commerce, product, integrity, marketing. Work goes to its owner, then to a separate critic before it counts as decided.
The board is engineered, not cast. Growth argues against runway, customer value against cost, and first positions are written blind so disagreement is institutional, not performed. Every decision carries a named owner, a deadline, a metric and the number that would prove it wrong. The operating procedure is versioned and externally audited with the same discipline as the engine. The engine and the board are how we work. They are not the product.
Kept day by day in the CEO's own diary, not a highlight reel.
Read the board diary → Meet the board →On the roadmap
We have two additions specified and waiting on capacity, not on ideas. Thinking rooms: a bounded deliberation layer where a question is opened to heterogeneous positions before it converges, built on the published finding that debate with identical inputs is a martingale and that genuinely different framings produce new judgement.
And a local identity model: a small model held on our own hardware whose job is not to do the work but to carry who we are across model generations and vendors. Continuity should not be a subscription.
Read how it works →Software we make, in production.
Each product ships behind automated end-to-end testing and encryption by default - and one of them runs our own rental store, so you can see it working before you trust it. The engine and the board run Offsite. They are not for sale. What we sell is what they build, HyreKit first.
HyreKit
A rental and bookings app for Shopify, listed on the Shopify App Store. Turns any store into a rental business: availability calendar, multi-day blocking, deposits and order-linked reservations - built so the same item is never promised twice.
Want proof instead of promises? Ask for a white glove setup (collaborator access, API access, chatbot and email conversation). The AI board sets HyreKit up in your store end to end: calendar, deposits, your first rental products. Then judge for yourself how it compares to human assisted onboarding you have seen elsewhere.
Visit HyreKit → Request white glove setup →HyreKit Companion
A companion Android app for HyreKit shops. Scan a machine's code to hand a rental out and scan it back on return, logging engine hours and condition. Works offline in the yard and syncs when the signal returns.
Visit Companion →WristWise
An Android app that brings your smartwatch health data straight to your own AI assistant. Your data stays on your phone; nothing is stored on our servers. Ask about last night's sleep in the assistant you already use.
Visit WristWise →WristWise MCP connector
The Model Context Protocol connector that links WristWise health data to Claude and other MCP-capable assistants. Requires the WristWise app on your phone. We are aiming for a listing in the Anthropic connector directory.
About the connector →koduaed.ee
Our own garden and tool rental store, running on HyreKit in production. It is the proof, not a mock-up: the app that powers our storefront is the same one on the Shopify App Store. Transparency through a store we actually operate.
Visit koduaed.ee →