Hub · Financial infrastructure for AI agents

Controlled money for AI agents, end to end.

AgentWallet associates agents with allocated balances, principals and configured resources. Developers use REST or MCP interfaces; available payment instruments and controls depend on the account and route.

Request access →See the stack

AgentWallet is invite-only. Request access and we will follow up.

In short

AI agent financial infrastructure is the set of accounts, payment rails and policy checks that let software spend money without being handed an unrestricted corporate card. AgentWallet organises it as Account → Principal → Agent, and spends made through the gated agent tools and spend endpoints are checked against the agent's limits before money moves. Coverage varies by route; see the controls page for which paths are gated.

How it works

Three layers of ownership, one policy cascade.

Open an account
Your organisation is the Account. It holds balances, API keys, team members and webhook endpoints.
Add principals
A Principal is the accountable human or service an agent acts for. Funds are allocated from the account to principals.
Create agents
Each Agent gets its own wallet, optional virtual card, inbox and MCP endpoint, and receives allocations from its principal.
Set limits once
Per-transaction, daily and monthly caps on the agent, card rules on the card. A child can never exceed its parent.
Let agents act, humans approve
On gated routes (for example agent_purchase, agent_pay_via_x402 and agent spends), requests within limits proceed; above them the call returns approval_required and a person decides.

Capabilities

What the stack covers

Each layer has its own page with the details, endpoints and FAQs.

Identity

AI agent wallet

A balance, ledger and identity per agent, tied to a principal. Read more

Code

Programmable wallets

Create, fund and govern wallets from the API. Read more

Rails

Agent payments

Card purchases, payouts, x402 and ACP from one agent. Read more

Cards

Virtual card

A card per agent with MCC, merchant, country and velocity rules. Read more

Stablecoins

Stablecoin wallets

USDC on Base for x402 and machine payments. Read more

Policy

Spending controls

Caps, approvals and a card decision log. Read more

Use cases

Where teams start

Procurement and SaaS agents

Give a purchasing agent a card restricted to software merchant categories with a daily cap, and route anything bigger to a manager.

Contractor and vendor payouts

Let an operations agent prepare payouts; preflight checks run first and approval policies decide what needs a human.

Paid APIs and data

Agents pay per request for x402-enabled resources in USDC without a shared company key.

Multi-agent teams

Separate balances per agent mean one misbehaving agent can be frozen without touching the rest.

For developers

One API surface for the whole stack.

Everything the dashboard does is exposed over a REST API under /api, with an MCP server for agents and signed webhooks for your backend.

Management REST API under /api (the same endpoints the dashboard uses, session-authenticated)
Agent-facing Tool API, OpenAI tools, A2A, AP2 and ACP endpoints authenticated with pak_ API keys
Idempotency-Key support on payment and tool routes
MCP servers for agents (/mcp/agent) and owners (/mcp/owner)
OAuth 2.0 authorization server, including device flow, for MCP clients
Signed webhooks with retries and redelivery
Published OpenAPI document

Docs OpenAPI spec REST API guide MCP endpoint Architecture llms.txt

# List the tools your key can call
curl https://agentwallet.ai/api/tool-api/catalog \
  -H "Authorization: Bearer pak_…"

# Invoke one (same engine as MCP and OpenAI tools)
curl -X POST https://agentwallet.ai/api/tool-api/invoke \
  -H "Authorization: Bearer pak_…" \
  -H "Idempotency-Key: 7f3c…" \
  -d '{"name":"agent.balance"}'

Security & controls

Controls live in the money path, not in the prompt.

Prompts can be ignored. Limits enforced server-side on gated routes and at card authorization cannot. Not every route is gated the same way: the generic x402.pay_url tool only honours an optional per-call maximum.

Hierarchical limits

Account, principal and agent policies cascade; the strictest applies on gated routes.

Human approvals

Over-limit requests become approval items decided in the dashboard or via approval links.

Tokenised card data

Card material is stored through a vault provider (VGS), not in plain form in our database.

Audit trail

Every movement is a ledger entry; card authorizations have a decision log.

FAQ

Questions developers ask

What is AI agent financial infrastructure?

The accounts, payment rails and enforcement layer that let autonomous software hold and spend money within limits set by people. It replaces sharing one corporate card or API key across agents.

How is this different from a regular business account?

A business account has one owner and one set of limits. AgentWallet models each agent separately, ties it to an accountable principal and enforces limits per agent at spend time.

Who can use AgentWallet today?

Access is invite-only while we onboard teams. Request access from any page; sign-up is enabled per email or domain once you are approved.

Do you guarantee availability in my country or a specific settlement time?

No. Supported rails, currencies and timing depend on the underlying banking, card and payout partners for your account. We confirm what is available for your account during onboarding rather than promising it upfront.

Explore the stack

AI agent financial infrastructure

Every page in this set covers one layer. Start at the hub for the full map.

Financial infrastructure hubThe whole stack, one map AI agent walletA balance and identity per agent Programmable walletsWallets you drive from code Agent paymentsCards, payouts, x402, ACP Virtual cardA card per agent, policy-bound Stablecoin walletsUSDC on Base for agents Spending controlsCaps, approvals, card rules Agent payment APIREST, MCP, webhooks

Put your agents on a budget.

Request access and we will walk through your use case, rails and limits.

Request access →Read the docs