A raw wallet cannot express purpose, budget, provider trust, or when a person must ask.
The financial control plane for AI agents
Give agents financial capabilities.Not unrestricted wallets.
Set budgets and approval rules once. Give agents a controlled path to paid tools and onchain actions.
Ordinary agent sample task
Sample payment and DeFi data. No funds move.
Ready to call
Find a lower-risk place for 500 USDC on Base. Spend up to $0.10 on current evidence. Prepare the action, but ask before execution.
please.plan(goal: "earn on 500 USDC", research_budget: 0.10, execute: false)The agent knows the financial outcome. It still needs paid evidence and a DeFi route.
please handles both under one task policy.The model requests. Policy decides. Every outcome is recorded.
Keep the agents you already use
One control plane. Bounded financial reach.
Ordinary agents can reason about outcomes, but they should not hold unrestricted spending authority. please gives each agent only the budgets and capabilities its operator defines.
Policy binds the budget and route, asks when required, and records the complete outcome.
From agent request to controlled outcome.
Request
The agent asks for a paid capability or onchain outcome without supplying authority.
Decide
Deterministic policy checks identity, purpose, provider, route, budget, and exposure.
Authorize
Bounded requests proceed; material, unfamiliar, or changed requests ask a person.
Account
Denied, purchased, prepared, submitted, and completed outcomes enter one audit trail.
One request. Paid research. DeFi action.
please can buy current evidence under a small budget, compare DeFi routes, and turn the result into an action preview that follows a separate approval rule.
View DeFi exampleCompare current USDC opportunities on Aave and Morpho on Base. Spend no more than $0.10 gathering evidence. Do not transact.
Aave market data$0.018
Morpho vault data$0.014
Risk and freshness checks$0.009
Total research cost$0.041
Authority stays outside the model.
Set what each agent may buy or do onchain, how much it may spend, which funding source it may use, and when it must ask.
- 01
Up to $0.01 per research call
Allow - 02
Up to $0.10 per research task
Allow - 03
Known research providers only
Allow - 04
No recurring charges
Deny - 05
Manual approval before any onchain action
Ask - 06
Deny if the recipient or action changes
Deny
Coverage by release state
One authority model. Many financial rails.
Agents use one please connection. Identity, policy, budgets, payment rails, DeFi protocols, routing, and receipts stay underneath.
Agent interface
One stable surface for the agent.
- MCPCurrent read-only interface
- A2APlanned
Paid capabilities
Discover paid services, normalize terms, and retry after payment.
- Coinbase BazaarIn development
- x402In development
- MPPIn development
DeFi action preview
Compare, route, and simulate supported onchain actions without moving funds.
- Aave V3Current read-only preview
- Morpho vaultsCurrent read-only preview
- Swaps and bridgesCurrent read-only preview
Authorization
Apply rules and preserve evidence of authority.
- please policiesIn development
- AP2 mandatesIn development
Commerce and metering
Track checkout state, catalogs, service credits, and usage.
- UCPPlanned
- ACPPlanned
- NeverminedPlanned
Identity and settlement
Bind agent identity and select a compatible route.
- Visa TAPIdentity layer
- Supported stablecoin railsPlanned
Payment support is not marked available until payment, retry, delivery, and receipt verification pass the release gate. Current DeFi support is read-only preview only.
Readable by people and agents
Every payment and action explains itself.
Evidence purchased, DeFi route prepared, policy applied, approval required, and remaining budget stay in one normalized record.
Exa Search was purchased for $0.007 as part of a $0.041 sample research task. A read-only DeFi action preview was prepared. No transaction was submitted.
- Research provider
- Exa Search
- Capability
- Primary-source web search
- Purpose
- Compare USDC opportunities on Base
- Research call
- $0.007 USDC
- Payment route
- x402 sample route
- Payment authority
- Research calls up to $0.01
- DeFi action
- Supply 500 USDC
- Action route
- Aave V3 on Base
- Action status
- Read-only preview. Not submitted
- Action authority
- Manual approval required
- Request digest
- req_84ac...19f2
- Result digest
- res_321b...7d0e
- Recorded
- 16 Jul 2026, 14:32 IST
- Research budget left
- $0.059
Add one MCP. Keep financial authority outside the agent.
please gives existing agents a stable interface for bounded paid capabilities and DeFi actions. The agent requests the outcome; deterministic policy controls the authority.
Join pilotAuthenticated payment access is limited to the pilot. The public MCP is read-only today.
Money movement is a deterministic system decision.
Natural language can request the outcome. Explicit rules decide payments and actions.
- The language model may request a purchase or DeFi action. It cannot change policy.
- Tool output is untrusted data and cannot authorize spend or asset movement.
- Payment recipient, DeFi route, amount, request, and expiry are bound before approval.
- Payments and onchain actions use separate authority and idempotency boundaries.
- Changed prices, recipients, routes, or action terms fail closed.
- The current public MCP remains stateless and read-only. It cannot move funds.
Give agents capability without unrestricted authority.
Start with one agent, one small budget, one bounded capability, and one readable history across supported payment and DeFi protocols.
