Documentation

A company's working software — CRM, ERP, processes, goals and knowledge — with an agent that can use all of it.

The everyday work of a small service company — customers, projects, processes, goals and documents — with an agent that can reach all of it.

Where to download it, what to expect on macOS and Windows, and what to do about the warning Windows shows.

Three or four short steps to your own workspace — or one button to look around with demo data first.

kyte is installed and the workspace exists — here is what to do in the first hour so it stops being empty.

Which version you are running, how kyte updates itself, and what the number means.

The chat is where the work happens — two modes, your permissions, and a switch that decides how far it goes alone.

The switch under the message box decides how far the agent may go on its own — from planning only, to acting without asking.

What counts as consequential, what the approval card shows you, and what happens when you say no.

Send a long job away with /parallel, keep working, and pick the result up when it is done — or stop it.

Which files the agent can read, what happens to them, and why they are not stored anywhere.

Give the agent a folder on your disk, and it can read, search and write files in it — inside that folder and nowhere else.

kyte spots names, addresses and account numbers in what you are typing, and lets you replace them before the message leaves the machine.

Pick the model for a conversation, see when it runs on your own network, and understand who decides what is on offer.

The agent keeps notes on how you work — you can read them, correct them, pin them or forget them.

Speak your request into the message box — useful for the long ones, which are usually the better ones.

Five apps on one database — a customer in ERP and an organisation in CRM are the same company, not two records that drift apart.

Who you do business with, who you talk to there, and what is currently on the table.

What you are delivering, who is on it, how long it took, and what that is worth.

The steps your company repeats, written once — then every run is its own checklist, done by a person or by the agent.

What your company knows, as markdown files — either in kyte or in a folder you picked, where the file is the truth.

What the company is working toward, broken into measurable results, hung on the people and teams responsible.

Things to be done, kept where the agent can see them — and deliberately not a queue it works through on its own.

Describe something your company needs, and kyte builds it as a working app — tables, forms, lists, permissions.

From two sentences to a working app — what to say, what you get, and what to do when it is not quite right.

Add your own fields and lists to CRM, ERP or Processes — beside what shipped, never inside it.

Ready-made descriptions — expenses, helpdesk, fleet — that Extend turns into an app in your workspace.

Three ways to hold a workspace's data — and the one thing they have in common: the app always reads and writes locally.

How someone joins the workspace, what a role decides, and why the agent inherits it.

On your machine first, encrypted when it belongs to a team — and what actually leaves it.

kyte can hand its reading tools to Claude or another AI program — as you, with your permissions, kept current by itself, and unable to change a thing.

Everything local keeps working; only the model and the sync wait for the network.

How to report it, what kyte reports by itself, and the two buttons worth knowing before you need them.