Owner-controlled engine · v0.4
Build the team behind the answer.
AI Stack turns model choice into a visible operating design: who works, in what order, with what evidence, under which data boundary, and where a human must decide.
Explore the architecture ↓STACK rolespolicymodelstrace
01 · Composition
Eight patterns, not one vague “agent mode.”
Each choice changes actual execution. Select a card to see the flow.
02 · Roster
A broad cloud bench. A deliberate local bench.
Swipe horizontally; the operating studio uses the same compact model-rail philosophy.
03 · Boundary
“Blocked” is an action, not a badge.
Public / non-sensitive
Enabled cloud models may run when their independent account and data terms are acceptable.
Confidential / regulated
The engine refuses the run before contacting any cloud provider. Only the selected local model is eligible.
Honest limit
The operator chooses the classification. This guardrail is not automatic data classification or regulatory certification.
04 · Ownership
The brain lives with the owner. Projects connect without surrendering themselves.
No credentials, prompts, client data, or live runtime state are published with this page. The deployed website explains the system; the local loopback studio operates it.
05 · Integration lab
One private engine. Many explicit clients.
The experiment is a versioned handshake—not a public AI endpoint.
GET /capabilitiesVerify AI Stack, schema 1, API v1, and loopback transport.
GET /stackRead the active secret-free workflow design.
POST /runsShow the exact packet before one human-approved request.
If AI Stack is running on this Mac, the first button opens its Studio. If it is not running, the browser fails locally—there is no cloud fallback and this website receives nothing.
Deliberate wall: samuelplawton.com can explain or launch the local surface, but it cannot read the engine or send it work. AI Stack emits no cross-origin permission and rejects cross-site mutation.