Section B• Last verified November 2026

Understanding Agents

Agent identities, capability registry, task delegation, and execution boundaries.

What is an Agent?

In Mandate, an Agent is a specialized autonomous worker equipped with a defined role, tools, and execution capabilities. Agents can analyze data, browse web resources, formulate DAG plans, query APIs, and propose financial disbursements.

However, agents in Mandate are strictly unprivileged proposers. They never possess raw cryptographic signing keys and cannot execute any transfer without an active, policy-compliant mandate signed by the user.

Capability vs Financial Authority

Crucial Distinction
An agent may possess the technical capability to interact with the NEAR blockchain, but it has zero financial authority unless an explicit, active mandate has authorized it to draw against a defined budget.
AspectCapabilityFinancial Authority
Defined ByAgent Capability Registry & ToolsUser Mandate Policy
Examplenear_transfer, web_searchMax 0.5 NEAR to researcher.testnet
TransferabilityAgents can sub-delegate tasksNon-transferable; bound to mandate ID

Agent Identities & Keys

Every agent in Mandate receives a tenant-isolated identity consisting of:

  • Agent UUID: Unique identifier in the PostgreSQL database.
  • Dedicated Implicit Account: A 64-character hex NEAR address derived inside OutLayer Intel SGX TEE.
  • Tenant Ownership: Strictly tied to the user's authenticated account session.

Agent-to-Agent Delegation & Swarms

When executing large objectives, a lead planning agent can assign tasks to specialized sub-agents (e.g. a research-scout delegating data compilation to an analyst-agent).

Delegation transfers work, never financial custody. Any financial proposal originated by a sub-agent is evaluated against the top-level mandate authorized by the human user.