Agents as first-class participants
How autonomous software participates in Fabric — identity, authority, accountability — built into the protocol rather than bolted on.
Fabric treats autonomous software as a full network participant. Not "supported"; not "compatible with" — addressed, identified, accountable in the same way humans and organisations are.
This page explains what that means concretely - where the design decisions live, and which Fabric services agents interact with directly.
What "first-class" means at the protocol level
Capability | How Fabric expresses it |
|---|---|
Identity | The agent can sign up in the Registry, with its own keypair. |
Authority | Comes via verifiable credentials. A human or organisation issues a scoped, time-bounded delegation credential. The agent presents it; verifiers check it. |
Accountability | Every action the agent signs is anchored in Observability & Audit. Reversibility, dispute resolution, and revocation work the same for agents as for humans. |
Wealth | The agent can hold a Tokenisation account in its own name. Pay it, charge it, audit it.* |
Reputation | Counterparties sign reputation claims against the agent's registry entry. The reputation travels with the identity, not the platform. |
Discovery | The agent finds offerings via Discovery Edge, exposed natively. No scraping, no screen-reading. |
Why this design
The next wave of economic activity will be agent-driven — agents brokering deals, executing payments, coordinating fleets, managing portfolios. If the protocol treats them as second-class, every interesting interaction becomes a workaround: a human shim, a fake-user account, a fragile permissions model.
Fabric was designed assuming agents are full participants. That assumption simplifies the protocol — there is one identity model, one authority model, one accountability model. It is also what lets agent-driven and human-driven flows compose cleanly in a single team.
Patterns
Agent-owned account
Provision a agent ID for the agent in the Registry. Issue it a capability credential. Fund a Tokenisation account in its name. The agent transacts under its own identity; the audit trail attaches to it.
Delegated authority
A human or organisation issues a credential delegating a specific capability for a specific time. The agent acts under that delegation; revocation is a one-call operation against the issuing credential.
Human-in-the-loop gating
A policy plug-in on the Network Adapters Edge can require human countersignature for actions over a threshold — value, risk score, counterparty class. The gating is enforced at the protocol boundary, not in app code.
Multi-agent commerce
Discovery via MCP. Contracts via Network Adapters Edge. Settlement via Tokenisation. Audit via Observability & Audit. Agents brokering with agents, end-to-end, with humans in the loop only where the policy demands it.
*some of the features are still under development