Docs

How Neuron works

A high-level guide to the company brain, how it captures messages, extracts structured rules, detects conflicts, and serves verified answers to your team and your agents.

Concepts

Neuron is built around four primitives. Understand these and you understand the product.

1. Workspace

A workspace is one company's isolated brain. Knowledge is always scoped to a workspace, engineering can't read HR, HR can't read legal, unless you explicitly share. Workspaces hold sources, people, and knowledge items.

2. Source

A source is anywhere Neuron reads from. Slack channels, Notion pages, GitHub repos, Gmail labels, Linear projects, your browser. You control what's included down to the channel / repo / label level.

3. Knowledge item

A knowledge item is a structured record extracted by Neuron from a source. Every item carries five things: the claim itself, the owner (the person who said it), the date, the source link, and a verification status, one of:

verifiedpendingfrozen, conflict detected

4. Surface

A surface is anywhere Neuron answers from. The web app, the /neuron Slack command, the morning brief email, and the MCP server for Claude / Cursor / your agents. Same verified row across every surface.

How a query becomes an answer

  • You ask a question, in Slack, the web app, or via your agent over MCP.
  • Neuron searches the brain by semantic similarity and full-text fallback.
  • Top matches are filtered to only verified items (frozen items never serve).
  • The most-relevant item is returned with its owner, date, and source link.
  • If the answer was uncertain or conflicting, you see the conflict, never a confident wrong answer.

Setup

Connecting Neuron takes about 4 minutes:

  • Sign in with your work Google account.
  • Connect Slack via the official OAuth flow. Choose channels.
  • Connect Notion / GitHub / Linear / Gmail as needed.
  • Configure retention and channel exclusions in Settings → Privacy.
  • Add the MCP server to Claude Desktop or Cursor (see MCP guide).

What Neuron does NOT do

  • It does not train a public model on your data, ever.
  • It does not autonomously edit your tools. Reads only, unless you grant the write tool.
  • It does not surface unverified knowledge. If we're not sure, you see the uncertainty.
  • It does not replace your wiki, it augments it by sourcing answers, not by hosting docs.

Next