Regent Protocol Start building
Platform

The complete system, before the details.

Regent is the financial authorization and compliance layer for AI agents that move money. Five primitives, one decision path — here is how they fit together.

System map

Where Regent sits.

Your applicationagents + business logic Regent control layer AgentID · Mandates · Guardian decides before execution SettlementSolana programs (devnet) Audit evidenceMerkle batches · Celestia DA Control Panel — human operator authorize anchors evidence
Your application keeps its business logic; Regent decides and evidences. Chain layers are network characteristics, separate from Regent application performance.
The primitives

Five answers to five questions.

AgentID

Identity →

Who is acting — a verifiable identity connected to a responsible party.

Mandates

Authority →

What it may do with money — amount, destination, time, conditions, approvals.

Guardian

Risk →

When to stop — rules and behavioural signals: allow, hold, or stop.

Audit Chain

Evidence →

Proving it later — hashes, Merkle batches, chain anchoring.

Control Panel

Oversight →

Human operators — monitoring, investigation, intervention.

Lifecycle

Before, during, and after a financial action.

Before

The agent holds an AgentID bound to a responsible party and a mandate bounding its authority. Nothing else runs without these.

During

Each action is authorized against the mandate and evaluated by Guardian — allowed, held for a human, or stopped. The deny happens before money moves.

After

The decision is hashed, batched, and anchored. Anyone can verify the record against the chain — without trusting Regent or the operator.

FAQ

What Regent is — and is not.

Is Regent a payment processor?
No. Regent authorizes and evidences; money always moves on your existing rail. We decide whether an action is within its mandate — the charge itself stays with your PSP, exchange, or wallet flow.
Is every API call recorded on-chain?
No. Financial-agent decisions that need proof are hashed and Merkle-batched; the batch root is anchored on-chain. Payloads live in the audit store — what is committed where is documented per artifact.
Where do humans fit?
Everywhere it matters: a mandate can require approval above a threshold, Guardian can hold an action for review, and the Control Panel carries the kill switch — revocation suspends the agent's mandates in under a second in our benchmarks.
What is available today?
Agent registration, mandates, authorization, and audit anchoring run live on Solana devnet, with a Python SDK and REST API. Mainnet, independent audit, and SOC 2 are on the roadmap — statuses live on the Trust Center.
Start building Apply for a pilot