How Synonode works

#The four parts

Part	Where	What it does
The app	Any browser or phone	Where you chat, see the preview and change settings.
synonode.com	The cloud	Serves the app, logs you in, and passes messages between the app and your runner.
Synonode Runner	Your computer	Keeps your projects and chats, runs commands and the browser, checks the work, and talks to the model.
The model	Your computer, or an API you choose	Does the thinking.

#No open ports, no tunnels

The runner connects out to synonode.com, the way a chat app does. Your computer needs no open port, no router settings and no tunnel. When the runner stops, the app shows which computers you have, which projects are on each, and how to start them again.

#Where your projects live

A project and its chats live only on the computer that has them. synonode.com keeps a list of which computer has which project, nothing more. When you open a project on another of your computers, it's copied there straight from the one that has it, through your account, and stored nowhere on the way. See Several computers.

#What leaves your computer

• Your messages and the replies pass through synonode.com between the app and your runner. They aren't kept there.
• Web searches and pages, when the model looks something up. You can turn web access off in Settings → Privacy.
• Requests to an online model, only if you set one up instead of a local model.
• Services you connect, when the model uses a connector.

#Your account

Accounts are approved by an admin before they can be used. The runner and its install command are only for approved accounts. Each runner logs in with your account, so it only ever works for you.

#How a turn works

When you send a message, the runner puts together what the model needs: Synonode's instructions, the project's instructions, what it remembers about you and the project, and the conversation. The model answers, often by using tools: reading and writing files, running commands, opening pages in the browser, searching the web or starting agents. Each step shows in the chat as it happens, and the final answer tells you what changed for you and how it was checked.

The model has no clock between your messages, so Synonode tells it what changed since your last one. When a new day starts, it gets the date. When you come back after a break of 30 minutes or more, it learns how long you were away, so it checks the preview, the files and anything it left running before relying on them. When a background job ends and wakes it, it's told this isn't a message from you, so it never takes a finished test run for your "yes". It also knows which model it is (the file llama.cpp serves, told again if you change it), so it doesn't answer "I'm ChatGPT". And when one of its own commands changes a file it read, it's told to read it again before editing it. You don't see these notes in the chat.