fleeTau is conceived, programmed, deployed and promoted by a single founder. The roles a whole department is usually hired for are performed by hundreds of AI agents. The human stays where a decision costs money: task setting and acceptance.
The product runs in production. Below — exactly what, and on what.
What we build, what matters more, and which finished work to accept.
1 person — the only one who decidesOn the left — how a product of this size is usually staffed. On the right — how it is staffed here. Not a single line is skipped: all the work gets done, only the executor changes.
The savings are not the point. The point is that between “decided” and “live in production” there are no approvals and no context handoffs left: it is the same person.
We do not “apply AI” — we run the company with it, and we sell fleets exactly the same loop. The product is proven on ourselves before any customer.
Breaks down the task, writes code, migrations, tests and copy, and presents the finished solution as a whole.
Sees idle time, an overdue payment, an expiring insurance policy — and proposes an action with the amount.
The founder accepts or rejects. The decision stays with him — and only the decision.
The fleet owner confirms from the phone with a single tap. The decision stays with them.
The pipeline builds, deploys, and verifies that the service answers from production.
A debit, a message to the driver, a charge to the partner, a record in the history.
From this follows what usually has to be promised: speed. Between a hypothesis and a working feature there is no sprint and no task handoff — there is task setting and acceptance.
Everything listed is answering requests right now — on a single machine in a Kazakhstan region, with pipeline deployment and one-command rollback.
accounts, people, vehicles, rental, messages, payments and the gateway
the site, the fleet workspace, the driver workspace, the partner workspace
embedded in the binary, applied under a lock at startup
a reference of 176 makes, a snapshot of the Kazakhstan market
Only the gateway has a public address. The other services have none at all.
Each service has its own database and its own queue: another service’s data cannot be read even by accident.
7 GB of memory, 20 GB of disk. That is what the entire infrastructure costs right now.
Releases sit side by side; switching the symlink is instant.
In a typical company a round turns into hiring, and the first months go to onboarding. Here the expense side does not grow with the workload — only the agent bill grows, and it is tied to work done, not to the month.
An order of magnitude, not a report. Exact numbers — in a conversation, with a data export.
That is why decisions do not live in one head. Contracts between services, role rules, breakdowns of expensive mistakes and the reasons behind every non-obvious choice are written down in the documentation repository — both the agents and the human read it. Whoever joins the project gets not access to people but a working system and its explanation.
Human acceptance is mandatory, and it is not the only line of defense. The database holds the boundaries: composite foreign keys, time-based exclusion constraints, uniqueness within a fleet. An agreement can be forgotten; a database constraint cannot.
Support is the first candidate for handing over to agents, and the product is already built for it: a messaging service, workspaces and a single event loop. A fleet asks right where it works.
A model is tested by speed over the long run, and it is measurable: every service reaches production through the pipeline, every deployment leaves a trace. We are ready to show that history in full, not a showcase.
Fifty minutes: live workspaces, the deployment pipeline, the decision repository, and how a task travels from statement to production in a single evening. Amount and terms come after — first it is worth making sure the model is real.
The founder answers. There is nobody else in this company to answer.