Zetaiti Platform
A hub where AI agents do real work — under your rules
Zetaiti Platform runs on your own machine. You talk to it like an assistant; its agents do the work — coding, designing, the paperwork in between — on the model you choose. What they do is visible as it happens, and anything that governs them waits for your click.
People
- Browser panel
- Android phone
- Wear OS watch
- Windows desktop
- Any computer
Zetaiti Platform
- Agents
- Approvals
- Permissions
- Memory
- Logs
Your world
- Machines and files
- Data, ERP and MES
- Your services and tools
- Models, local or cloud
How it fits together
People, the hub, your world
One hub on a machine of yours. People reach it from any device; its agents reach what you allow.
People and devices
A panel in the browser; apps for Android phones, Wear OS watches and Windows desktops; and a small client that lets any Linux, macOS or Windows computer join. A device pairs with a QR code and answers to its person.
The hub
Agents, memory, approvals, permissions and logs, on one machine of yours. HTTPS only; by default reachable from that machine and your private Tailscale network, nothing more.
Your world
Your machines, files and services; models running locally or at a provider; tools connected through the open Model Context Protocol (MCP).
What it does
Built to work, built to be watched
Agents that do the work
An Orchestrator takes your request and hands the work to work chats, specialists and teams. They write code, design, handle documents, and use computers of their own — isolated machines where they can install and try things safely.
Any model, local or cloud
Open models on your hardware through Ollama, llama.cpp or vLLM, or any OpenAI-compatible provider — OpenAI, Anthropic, Google, Mistral, DeepSeek, OpenRouter and more. Tools, skills and specialists give a mid-size model the structure for serious work.
Approvals that fit the risk
In Auto mode the agents work without interrupting you; in Manual mode you are asked about every call that does something. Signing in, paying or submitting is asked in every mode.
Permissions per person
Viewer, member, admin, owner — or a level of your own — with grants for recorded exceptions. The agents working for a person act within that person's level.
Everything visible
The Workstream shows what the agents are changing while they change it; logs, traces and a chronicle keep who did what. It is all built from events in fixed words, never from a model's summary.
Every device is part of it
Phones, watches and computers join the hub and lend it what they have — files, a screen, a shell — always held to the rights of their person.
Talk to it
Type, or speak: voice calls with the agent from the panel, the phone or the watch.
Learns, then repeats
Memory with rules you can read and edit. A task solved once becomes a skill or a recipe, so the next time is quick and the same; packs carry them to another installation.
Safe by default
Secrets are used, never shown to the agents. Backups are one encrypted file (AES-256), on demand or on a schedule, copied to any S3-compatible store if you wish. A new version can be switched back.
For industry
From the office to the shop floor
Zetaiti Platform is a management and assistance layer: it reaches the systems you already run and puts them within reach of people and agents.
ERP and MES
Integrations through connectors, APIs described with OpenAPI, and MCP servers — built for your setup by our engineering team.
Machine data
Data from controllers and PLCs brought in through integrations we build with you, read-only unless you decide otherwise.
Vademac inside
Vademac can run as a connected service shown inside the panel, its key held by the hub and never by the browser.
Where people already talk
Messages from Telegram, Matrix, Slack or email can reach the agents, and their answers go back the same way.
Integrations with a specific ERP, MES or controller are engineering work scoped with you; we do not sell certified off-the-shelf connectors.
Getting started
One command, then a guided set-up
On Linux, macOS or Windows, one command installs it and makes it start with the machine. A guided set-up looks at the machine and proposes only what fits — a model it can run, or a provider you choose.
- Linux · macOS · Windows
- Docker, GPUs and local models: optional — it uses what the machine has
- HTTPS only, from the first start
- Updates you can switch back
Want to see it on your machines?
We will show you the platform on a real task of yours, and what it would take to run it in your company.