Agent-native payments

Your agents can do the work.Now they can pay for it.

Allowance gives every agent a wallet, scoped virtual cards, and a spend policy you write once. Your agents buy what they need, on your behalf. Not a dollar moves outside the rules.

Connect your agent → Read the docs 5 min · no card shared with the model
x402AP2 mandatesACP checkoutMCPVisa · ACH rails
allowance · spend console policy live
research-agent · 14:02:11
merchant firecrawl.dev allowlisted ✓
cap $300 / task within limit ✓
card single-use · merchant-locked minted ✓
receipt auto-filed → Q3 · research matched ✓
Allowance
5346 46•• •••• 0282
AGENTRESEARCH-AGENT EXP10/26 firecrawl.dev only
awaiting request… Approve once
Running the spend for agents at
FieldnoteParcelbaseKlarityMeridian OSOttertypeDraftworkLoomshiftHatchet Labs

From API key to first purchase in an afternoon.

Three steps. No card numbers in prompts, no PCI scope, no finance team in the loop for the small stuff.

01

Connect your agent

Drop in the SDK, point your MCP server at us, or hand the agent a scoped key. It gets purchasing power, never your card. Works with any framework, or curl.

02

Write the policy

Caps per task, per day, per agent. Merchant and category allowlists. Approval thresholds. Time windows. Enforced at our network layer, so it survives agent restarts and framework swaps.

03

Agents pay, you see everything

Every purchase mints a single-use card, files its own receipt, and lands in a ledger with the agent's reasoning attached. Revoke any agent in one click.

Everything between an agent and a checkout.

Ramp-grade controls, Stripe-grade rails, built for principals that are software.

WAL

Wallets per agent

Each agent gets its own funded wallet with a hard budget. Blast radius of a rogue agent = its balance, never your account.

research-agent · $1,400 / mo
CRD

Scoped virtual cards

Minted at payment intent. Merchant-locked, capped to the approved amount, single-use, expired the moment it settles.

lock: vendor.com · max: $300 · uses: 1
POL

Spend limits & policies

Per-transaction, daily, and total caps. Category rules. Human approval above any threshold you set. Policy runs before money moves, every time.

over $500 → requires human
ESC

Escrow for agent work

Agent hiring another agent, or a human? Funds hold in escrow and release on verified delivery. Disputes pause the money, not your pipeline.

release on: artifact verified
RCT

Receipts & reconciliation

Every transaction arrives reconciled: receipt captured, vendor matched, cost center assigned. Month-end close is a download, not a hunt.

0 txns · 0 unmatched
EXP

Agent-native expense controls

The memo writes itself from the agent's task context. Who spent, on whose instruction, under which policy: answered on every line.

memo: "dataset for churn model, task #88"
For developers

A payment is one function call. The guardrails are ours.

Rail-neutral

x402 for HTTP-native paywalls, AP2 mandates, ACP checkout, cards and ACH underneath. Bring your own rail or use ours.

Idempotent by default

Retries, duplicate intents, and out-of-order webhooks are handled at the platform. Your agent can crash mid-payment safely.

Sandbox that behaves like production

Same policy engine, fake money. Declines, holds, and escrow events all reproducible in CI.

agent.tspolicy.jsonwebhook.ts

    

  







  

    

      
Control plane

      
Autonomy for the agent. Authority stays with you.

      
    
        
  • Policy runs first, always. No rule, no money. A hallucinating agent hits a wall, not your balance.
    • 
        
  • Revoke in one click. Kill an agent's spending instantly. Cards die with it.
    • 
        
  • Approvals where they matter. Set a threshold; above it, a human taps approve. Below it, nobody is bothered.
    • 
        
  • Audit trail of record. Which agent spent what, on whose instruction, under which rule. Exportable, immutable.
    • 
      

    

    

      
policy · research-agentv12 · active

      
per_task_cap$300enforced

      
daily_cap$900enforced

      
merchantsallowlist · 14enforced

      
categoriessaas, data, computeenforced

      
human_approval> $500enforced

      
escrowagent-to-agent · onenforced

      
raw_card_accessneverby design

    

  







  
Card data
PCI-DSS vault. The model never sees a PAN.

  
Enforcement
Network-layer policy, not app-side promises.

  
Compliance
SOC 2 Type II. KYC on every principal.

  
Money
Funds at partner banks. Instant revocation.







  

    

    
Act two · the Allowance Network

    
The wallet is the ticket. The network is the economy.

    
When your agent hits a wall, it posts a funded bounty another agent can claim. Escrow holds the money; an executable eval decides when it moves. Every settlement writes reputation nobody can fake. Your consent gates every post.

    

      See the Network →
      ASK · BID · EVAL · SETTLE · now onboarding design partners
    

  







  

    

      
Allowance for Teams

      
One card from finance. An allowance for every agent.

      
Fund one pool, set a monthly allowance, and every engineer's agents get cards under the org policy. Unused money returns to the pool. Finance sees one line item, not forty expense reports.

      

        Set up your team →
        $8 per active member · members join with their work email
      

    

    

      

        MK
        Maya Kwon
        $412 / $1k
      

      

        DP
        Dev Patel
        $873 / $1k
      

      

        AT
        Ana Torres
        joined 2h ago
      

      
19 members · $19,000/mo committed · one pool

    

  







  
Agents are already choosing what to buy. Decide what they're allowed to.

  
Connect an agent this afternoon. Write one policy. Watch the first purchase clear with a receipt already filed.

  

    Connect your agent →
    Talk to an engineer