Agentic
This section is under active development. New topics and guides will be added as the work lands.
AI agents and large language models (LLMs) are increasingly able to act on a user's behalf — browsing catalogs, filling carts, booking, and paying through tools exposed over the Model Context Protocol (MCP). As soon as an agent touches a real-world transaction, it runs into identity: proving age, applying a loyalty membership, or authorizing a payment. These steps can't be hallucinated — they need verifiable, cryptographically-backed claims.
This is where Multipaz fits into the agentic stack. Instead of the agent asserting "the user is over 21," the flow requests a digital credential and the user presents it from their wallet (selective disclosure, bound per request). The agent orchestrates the conversation; the credential layer provides the proof. The result is agentic experiences that can transact while keeping the human in control of their identity and money.
The CredentAgent SDK
CredentAgent is the consent layer for AI agents — the SDK that puts a credential check in front of an agent's consequential actions. Before an agent completes a payment, an age gate, or an access grant, CredentAgent makes it prove a verifiable credential from the user's wallet, enforced server-side. Its guiding principle: identity leads, and payment is just one application — age, membership, and payment are all just credentials in the same typed policy. It ships as two npm packages (a host-agnostic Gate and a ready-made Storefront) and runs across Claude, ChatGPT, and Goose.
Interaction models
How a person participates in the flow defines the interaction model:
- Human Present — a person is in the loop to approve the sensitive steps: presenting a credential from their wallet and authorizing payment (for example with a passkey / biometric). This is available today, demonstrated by the Product Picker reference app.
More interaction models will be documented here as they mature.