SELF-HOSTED · GOVERNANCE-FIRST · CLAUDE AGENT SDK

One core.
N instances.

A neutron star doesn't outshine the field — it out-binds it. neutron collapses a whole agent platform into one small, dense core, then holds N configured agents in orbit with gravity: roles, approval gates, and rules that never let power escape.

# one image, one database URL, one auth token, one instance.yaml
$ docker run neutron a running agent
1
image to run
N
instances / config
3
safety postures
0
code forks
01

Built like its namesake

The physics is the architecture. Everything a neutron star does with mass, neutron does with configuration.

Stellar physics

A star collapses into a core a few kilometres wide — matter at its densest.

Neutron core

The whole platform is a thin chrome around the Claude Agent SDK — a core small enough to read, dense enough to run everything.

Gravity binds it

Immense gravity holds the star together. Nothing escapes it.

Governance binds it

Approval gates, roles, and non-escalation rules hold every agent. No tool call, no config change, no privilege moves without them.

Pulsar beams

One spinning core sweeps many beams across space.

N instances

One image, deployed once, presents as many branded agents — each with its own persona, tools, channels, and repos. All of it is data.

Crust → outer core → inner core

A neutron star is layered — each shell a different state of matter.

Channel → engine → extension

Three seams, independently swappable. Chat surfaces, the agent engine, and your capabilities plug in — they're never welded in.

02

Gravity is the product

Most agent stacks treat governance as a safety valve. neutron treats it as the point — the binding force that makes autonomous agents safe to hand real access.

Gates

Approval cards

Risky tools stop at a human card in the chat — command, cost, and blast radius spelled out before anything runs.

Roles

Separation of duties

Approvals can require a role, and the requester never approves their own action. Approvers get a cross-thread inbox.

Invariant

No escalation

Agents can manage agents — create them, tool them, scope them — but can never grant more privilege than they hold.

Posture

Three modes

Read-only by default, approval-gated in the middle, read-write when you mean it. Per instance, per tool.

03

A real agent, in a chat you own

Token-streamed answers, a live activity trace of every tool call, rich markdown, and per-turn cost — running on your infrastructure, against your data.

neutron · your-instance
neutron chat — landing with capability cards
04

Meet it where the work is

The web UI is one surface. @mention the bot in a GitLab or GitHub comment and it investigates and replies inline — same memory, same gates.

Settings · Channels
Channels settings page
CHANNELS

GitLab & GitHub mentions

One conversation per merge request or issue, with full thread memory. Configure providers, allowlists, and write-only secrets entirely from Settings — no restart.

  • HMAC-verified webhooks, project allowlists, loop-safe
  • Self-hosted GitLab & GitHub Enterprise via base URL
  • Off by default; enable per instance
READY WHEN YOU ARE

Run your own star.

A container, a database URL, an auth token, and one config file. That is the whole install.

Deploy neutron → What's new