Skip to content

Everything is a Neuron

Jump to bottom
DigitalBrain Wiki edited this page Jul 16, 2026 · 4 revisions

Everything is a Neuron

Thesis: DigitalBrain does not compare things to neurons. It is a workspace of Neurons. Buttons, Login, Gmail, Salesforce, Chat, Features, Connections, Approvals — each is a Neuron identity you can name, read, drive, and (when needed) approve. Implementation differences are technical traits of the same kind of existence, never a second product taxonomy.

Why this is the product

If Login is a “screen,” Chat is a “session,” Gmail is an “integration,” and Features are “packages,” the product shatters into five mental models. Humans learn one UI; agents learn another; connectors become settings; approvals become a sidebar chore.

When everything is a Neuron:

  • One identity language — the same thing humans see and agents invoke.
  • One workspace — Owner/Actor scope, not a UI copy and an agent copy.
  • One trust story — external change is always an Effect Neuron, never a silent side effect of a clever agent tool.
  • One evolution story — new behavior enters as Neurons (often Features) through human-approved rail, not free shell.

Taxonomy of existence (not a second ontology)

Product nouns map to Neurons. How they are built varies:

You encounter… Exists as… Built how (trait, not category)
Login Login / Session Neurons Deterministic auth + durable session
Ask / Chat Chat + Operation Neurons State + optional model work + projection
Gmail inbox verbs Capability faces on a Gmail-backed Neuron Connector-backed queries
Salesforce update Effect Neuron awaiting decision Prepared mutation + approval
Google / Salesforce link Connector Neuron OAuth lifecycle + health
Feature / pack Feature Neuron (module + work) Approved programmable module
Activity row Activity / Operation projection Observation of work and Effects
Flutter destination Surface / UI Neuron Projection + bound actions
MCP tool call Same Neuron invocation, agent client Authenticated transport only

Do not invent product categories for “Agent,” “RFW surface,” “tenant shell Feature,” or “external MCP package.” Those are non-products (What We Delete).

What stays one identity

A Neuron has a stable address in the workspace (Owner/Actor space + Neuron id). Session transport, access tokens, Flutter routes, MCP connection IDs, and server replicas are caller context, not the product thing.

Two clients can hold different sessions and still resolve the same Chat Neuron. That is why MCP can light up Flutter: same address, same revision line, same surface feed.

Scenes (product feel)

  • You open Ask: you are looking at the Chat Neuron’s live projection, not a local message list Flutter invented.
  • An agent runs ino_interact (or later neuron_invoke on Chat): the Chat Neuron advances; Activity and Ask surfaces revise; Flutter redraws.
  • You connect Google: a Connector Neuron becomes healthy; capability faces unlock — not “MCP tools got installed.”
  • A Feature wants to write Salesforce: an Effect Neuron appears for approval; nothing external mutates without that moment.

Implementation is allowed to be boring

Orleans grains, pure functions, OAuth adapters, LLM budgets, and Flutter widgets are how. The product sentence stays:

This is a Neuron in the shared workspace.

flowchart LR
  subgraph Workspace["One Owner/Actor workspace"]
    L[Login / Session]
    C[Chat]
    F[Feature]
    G[Connector Google]
    E[Effect]
    A[Activity]
  end
  Human[Human via Flutter] --> Workspace
  Agent[Agent via MCP] --> Workspace
  Workspace --> Surfaces[Surfaces / projections]
  Surfaces --> Human
Loading

In one line: Everything addressable is a Neuron; traits change execution, never the product story.

DigitalBrain

Clone this wiki locally