What neutron is
Neutron is a web application you chat with. Behind the chat window sits an AI agent that can do real work: read your code, open a merge request, search the web, run a scheduled job, store a secret. It runs on your own servers.
One copy of the program can serve many separate organisations at once. Each one gets its own configuration file, its own database, its own users, and its own rules.
Everything in this guide hangs off that sentence. The important word is rules.
The one idea. Most AI agent tools assume the person talking to the agent is trusted. Neutron assumes they are not — or rather, that “trusted” is something you spell out per person, per tool, per action. An agent may read freely and still be blocked from writing. A person may be allowed to ask for a deployment and not allowed to approve their own request.
The difference, drawn
Where to go next
Every element in one table: what it is for, who uses it, where it is on screen, and which gate governs it.
What is actually running, and why there is one image rather than many forks.
The gate, the three safety modes, and why you cannot approve your own request.
How an issue becomes a merge request, and how far autonomy is allowed to go.
Thirteen journeys from connecting a vendor to answering a reply, with an architecture chapter behind them.
Managing your company’s work: a board, a week, a review gate, and agents that pick up what is ready.
One endpoint any agent can hold: connecting, every tool and the rule it keeps, and what a token can never do.
Neutron Cores federating with each other, each controlling its orbits centrally.
The six repositories, and how a change reaches the live system.
What it is genuinely good at, and the five gaps that are real.
Last updated on