Skip to content

How it works

Jump to bottom
DigitalBrain Wiki edited this page Jul 16, 2026 · 2 revisions

How it works

Short detail behind the Home story. Still product language — not an engineering manual.

One brain, two doors

Who Door Same brain?
You Sign in → app Yes
AI agent DigitalBrain MCP Yes

Different sessions are doors, not second copies of Chat or Gmail.

flowchart LR
  App["App"] --> W["Workspace Neurons"]
  MCP["Agent MCP"] --> W
  W --> Screens["Live screens"]
  Screens --> App
Loading

The app is a window

Screens (Chat, Features, Connections, Activity) project the brain. They do not own a private “real” Chat history or connection list.

If an agent changes something, the window should update. That is the product feel of co-presence.

Connections → abilities

  1. Open Connections.
  2. Connect Google / Salesforce / ….
  3. Health becomes Healthy (or Needs reauth / Disconnected / Misconfigured — be honest).
  4. Unlocked capabilities appear only when healthy.
  5. Features and Chat can use those abilities.
  6. Writes still need approval (next section).

Not: “install an agent package” or “paste a Gmail MCP URL.”

Approvals (trust)

Anything that changes the outside world:

  1. Brain prepares the change (clear summary).
  2. You Approve or Reject.
  3. Only then does the connection act.
  4. Activity shows the result.

New Features (skills) use the same spirit: propose → you decide → install.

Agents

Agents are clients. They may:

  • Drive Chat (same conversation you see)
  • Inspect / help with Features
  • Propose growth

They must not:

  • Own a shadow memory the app cannot see
  • Silently mutate Gmail/Salesforce
  • Become a marketplace of random tools

Growth

Propose a Feature → You decide → Install → Pause / Resume / Rollback if needed

That is how the product gets stronger without free-for-all automation.

In one line: Same brain, live window, connect then unlock, approve the outside world, grow on purpose.

Next: Demos · Back: Home

DigitalBrain

Clone this wiki locally