Regent Protocol Start building
Financial infrastructure for autonomous agents

Financial authorization for AI agents that move money.

Regent is the financial authorization and compliance layer for AI agents that move money: a verifiable identity for every agent, limits it cannot exceed, risk signals watched in real time, and tamper-evident audit evidence for every action.

Start building Apply for a pilot
Developer access available · production availability follows published status
The financial action flow
Agent
requests a financial action
AgentID
identity verified against the registry
Mandate check
amount, destination, time, approvals
Guardian
rule + behavioural risk signals
Payment rail
the action executes — or never does
Audit proof
hashed, batched, anchored
ALLOWED mandate OK · risk clear → executed · evidence anchored · visible in the Control Panel
Illustrative workflow — a conceptual walkthrough, not a live production integration.
Where do you stand

Three ways in, one control layer.

From SDK to a first verified action

Issue an AgentID, create a mandate, and authorize your agent's first governed action against it — then watch the audit evidence anchor. The quickstart walks the whole loop.

Read quickstart
The boundary

Access control answers what an agent can reach.
Money needs more.

Generic gateways help manage what an agent can access. Regent is designed to govern and prove what an agent is authorised to do with money.

Identity

Who is acting

A verifiable identity connected to a responsible party — not an API key that anyone could be holding.

Pre-execution limits

What it may do

Amount, destination, time window, and approval conditions are enforced before the action executes, not reviewed after.

Behavioural control

When to stop

Rule and risk signals can allow, hold for a human, or stop an action that is technically permitted but behaviourally wrong.

Evidence

Proving it later

Every decision leaves tamper-evident evidence that an auditor can check without trusting the operator.

The platform

Five primitives, one workflow.

Each one is a plain answer to a question a financial agent forces you to ask.

AgentID

Identity

Gives an agent a verifiable identity connected to a responsible party. Technical issuance is included; optional human verification is a separate, pass-through step.

Mandates

Authority

Defines what an agent may do with money: amount, destination, time, conditions, and approvals.

Guardian

Risk

Evaluates rules and behavioural risk signals, then allows, holds, or stops a risky action.

Audit Chain

Evidence

Creates tamper-evident evidence through cryptographic hashes, Merkle batches, and chain anchoring.

Control Panel

Oversight

Gives human operators monitoring, investigation, policy, and intervention surfaces.

Explore the platform →

Together they implement KYA — Know Your Agent: Regent's framework for connecting an autonomous agent to a responsible context, financial mandate, and verifiable audit history. KYA is Regent's product framework; it is not a formal regulatory standard or regulatory approval.

Identity

An account proves nothing outside its vendor. A document does.

An account in a vendor's database

Most "agent identity" is a row behind someone's login: credentials in a vault, permissions in a dashboard. When that vendor is down, gone, or simply not asked — it proves nothing. Nobody outside can check it.

A document the agent carries

Every Regent agent signs its requests (RFC 9421 — Web Bot Auth and AAuth dialects) and has a public registry entry: status, responsible owner, behaviour tier. Any bank, merchant, or platform can check it — without asking us.

POST /v1/trust/lookup · live registry 
Press a button — the check runs against the public agent registry, right now.

Lookups key on the RFC 7638 thumbprint of the agent's signing key. Behaviour tiers appear once a signed evaluation lands and vanish when it expires. The same registry powers Agent Validation.

Workflows

Built for the moment money moves.

Agentic payments

An agent initiates a payment. The mandate bounds amount and destination before execution; approval conditions can require a human; the evidence pack shows exactly why it was allowed.

Fintech platforms

A platform issues AgentIDs for customer-facing agents and applies account-specific controls — every customer's agent runs under that customer's limits.

Treasury & custody

Policies on transaction value, counterparty, account, time window, and multi-party approval — enforced before movement, evidenced after.

Compliance & risk

Case reconstruction from a single record: action → responsible context → policy → decision → evidence. Evidence packs are exports and mappings for internal review.

Status, plainly

What exists today. What is planned.

The public site is deliberately more precise than a pitch deck — every claim below carries its state.

Available now DEVNET

  • Agent registration, mandate registration, audit anchoring on Solana devnetLIVE
  • Python SDK — agents, mandates, audit, guardianLIVE
  • REST API with bearer-token auth and org scopingLIVE
  • Dashboard: agents, mandates, audit log, kill switchLIVE

On the roadmap PLANNED

  • Mainnet availabilityPLANNED
  • Independent security auditPLANNED
  • SOC 2ROADMAP
  • EU AI Act evidence mappings and exportsPLANNED
Developers

A first verified action in one sitting.

Register an agent, bound it with a mandate, authorize an action against it, and log the evidence — the full loop, from the real quickstart.

# pip install regent
auth = await r.payment.authorize(
    mandate_id=MANDATE_ID,
    request=AuthorizeRequest(amount=Decimal("25"), currency="USD"),
)                                     # the mandate says yes — or this raises

await r.audit.ingest_event(IngestEventRequest(
    event_id=f"action-{auth.jti}",
    agent_id=AGENT_ID,
    event_type="action.completed",
    payload={"authorization_jti": auth.jti},
))                                    # evidence queued for anchoring
Read quickstart Open docs
Trust & security

Security posture, not certificate claims.

No bearer secrets on agents; identity is proof-of-possession. Mandates fail closed — if the control layer is unreachable, the money action is refused. Evidence verifies against published keys, so an auditor does not have to trust the operator. Independent audit and SOC 2 are on the roadmap and will be stated here only when the reports exist.

Review security approach Contact security
FAQ

Frequently asked questions.

The essentials on Know Your Agent, spending controls, and agent identity.

What is Know Your Agent (KYA)?
Know Your Agent (KYA) is a compliance framework that verifies AI agents before they execute financial transactions. It ensures that every autonomous payment has a verified sender, approved limits, and a tamper-evident audit trail. KYA is Regent’s product framework; it is not a formal regulatory standard or regulatory approval.
How does Regent Protocol prevent unauthorized AI spending?
Regent Protocol uses spending mandates — programmable rules that define what an AI agent can spend, on what, and up to what limit. A transaction outside those rules is denied before it executes, and the agent never holds the payment credentials itself.
What is an AgentID?
AgentID is a unique decentralized identifier (DID) assigned to every AI agent in the Regent network. It links the agent to its KYA verification, mandates, and audit history.

Build with it.

Developer tier with devnet access. Issue AgentIDs, write mandates, authorize actions, watch evidence anchor — before you talk to anyone.

Start building

Put it in front of your risk team.

A pilot conversation with an architect: your workflow, where the mandate gate sits, and what the evidence pack contains for your reviewers.

Apply for a pilot