Skip to main content
The decision is not which transport. It is which surface of the one hosted platform your agent runs on. Every surface below is the same governed platform — see Overview — reached a different way.

​
The five surfaces

​
What each surface is

Admin Assistant

Org-authorized wallet and operational intelligence inside the WalletSuite console. WalletSuite operates the model and tools; the operator asks, the Assistant answers and acts within policy.

Embedded Assistant

The same Assistant API and templates, rendered white-label inside your product. Your users never leave your UI; the model, tools, and execution controls stay behind the scenes.

Customer-owned Agent

You run your own model or framework and connect to the hosted MCP — connect your agent.

Background Adaptive Agent

A webhook, API event, or schedule invokes contextual analysis with no chat surface. Same Assistant reasoning, triggered by an event instead of a prompt.

Deterministic Workflow

Fixed conditions and actions bypass the LLM entirely while reusing the same data and execution services. Use it when the logic is known in advance and no reasoning is needed.

​
Invariants across every surface

Chat is a presentation surface. It never grants authority. Whether a surface has a chat box or not is a UI choice. Authority is the agent’s five bound things. The policy engine decides on the intent, not on the surface it arrived from.
  • One organization runs many assistants and agents — not one per end user. You provision them once and route your users through them; you never mint an identity for every person who opens a chat.
  • assistantId is selected by your application code, never by the model. The model cannot choose which assistant it is. Your code binds the assistantId; the model operates only within that selection.
  • The agent’s key stays server-side. On the customer-owned surface only a short-lived token reaches the agent — see Connect your agent.

​
Which one fits

1

Operating your own org's wallets from the console

Use the Admin Assistant — it is already there, org-authorized.
2

Putting an Assistant in front of your own users

Use the Embedded Assistant — same Assistant API and templates, rendered in your product.
3

Running your own model or agent framework

Use the Customer-owned Agent and connect your agent.
4

Reacting to events with no human in the loop

Use a Background Adaptive Agent for reasoning on a trigger, or a Deterministic Workflow when the logic is fixed and needs no model.

​
Next steps