Your agent is using your card
Real credit cards and raw API keys end up pasted into agent context. One leak exposes the actual account — with no spend cap and no way to revoke just that agent.
Private beta · 2026
Give your AI agents a wallet you control.
Each agent gets its own spending limit, scoped credentials, and a hard stop that fires before anything is charged. Works with every payment rail and every LLM provider — so one runaway agent can't drain the company card.
- Drop-in SDK ·
pip install runvault - Hard spend caps, enforced atomically
- One integration, every payment rail
agent.py
# Give the agent a wallet with a $500 ceiling.
agent = runvault.connect(
agent_id="research-agent-01",
budget=500.00,
scope=["llm", "commerce"],
)
# Same identity. Any rail. No raw card number.
agent.pay(merchant="amazon.com", amount=29.99)
# ↳ RunVault picks the rail, issues a one-use card,
# enforces the cap, logs the spend.
$
0.00
/ $500.00
The problem
AI agents spend real money. Nothing was built for them.
Today most teams hand their agent a shared API key or a company card and hope for the best. When something goes wrong — a prompt injection, a bad loop, a compromised dependency — there's no per-agent cap, no audit trail, and no way to stop a charge before it clears.
Every checkout is a different integration
Stripe ACP, Visa VIC, Google AP2, x402 — six agent-payment protocols in 18 months. Each one needs its own registration, credentials, and merchant list.
Budget checks happen too late
Payment networks authorise after the transaction is submitted. Concurrent agents race each other and collectively overspend before any check fires.
LLM spend and purchase spend live in different systems
An agent can be well within its Claude budget while dramatically overspending on purchases. No single view of what an agent actually cost you.
A bad agent on one rail looks fine on the next
Reputation doesn't follow the agent. Merchants can't screen an agent before fulfilling a high-value order, because every rail has its own registry.
"Every financial system built in the last fifty years assumes a human is the one spending. Agents aren't humans. They need a wallet of their own."
How it works
One SDK call. Three things happen.
RunVault sits between your agent and the outside world. You write two lines; we handle the wallet, the cap, and the rail underneath.
A wallet for the agent
runvault.connect() issues a signed, short-lived credential
tied to one agent. You set the ceiling and the scope — LLM calls,
purchases, or both. Revocable in under a second.
A hard stop, before the charge
Every spend goes through a single atomic check. The cap is enforced before any upstream call — so a thousand concurrent agents can't race past the limit.
Any rail, same call
Virtual card today. Stripe ACP, Visa VIC, stablecoin tomorrow. RunVault picks the right rail for the merchant. You never re-integrate.
Why now
The rails are live. The wallet isn't.
In the last eighteen months, every major network shipped an agent payment rail — and none of them agree on who the agent is.
-
Visa opens its network to AI agents
Partners include Anthropic, OpenAI, Stripe, and Microsoft.
-
Stripe ACP, Google AP2, and "Buy it in ChatGPT" all launch
Three competing agent-commerce protocols in one quarter. Agent checkout becomes mainstream.
-
Visa Trusted Agent Protocol + Mastercard Agent Pay
Cryptographic agent authentication at merchant checkout. Cloudflare and Worldpay co-launch.
-
Perplexity + PayPal ships agent shopping for US consumers
Free agentic shopping. Agent-initiated purchases settle via PayPal.
-
Hundreds of real agent purchases in production
Six rails. Zero shared identity. That's the gap RunVault fills.
Market
Revenue scales with agent activity, not headcount.
$42B
LLM API market by 2026
Every AI team building agents needs per-agent cost controls.
$1.2T+
Agent-mediated commerce GMV by 2030
Visa, Stripe, Mastercard, and Google are racing to own the rail.
$4T+
Card GMV addressable via virtual cards
Works on every merchant today — no protocol adoption required.
$800M
Immediate SAM — AI platforms + multi-agent infra
Self-serve signup. No enterprise procurement cycle required.
Pricing
Start free. Pay when your agents do.
The SDK is MIT-licensed and free forever. Managed infrastructure, virtual cards, and the reputation network are paid tiers.
Developer
Free
For hobby projects and evaluation.
- Open-source SDK, self-hosted Redis enforcement
- Bring your own API keys
- Up to 5 active agents
- Community support
Startup
$49
– $500 / mo
For AI-native teams shipping real agents.
- Managed vault + atomic enforcement
- Virtual card issuance
- Standard rail adapters
- Shared reputation signal
- Audit log API
Enterprise
$2,500+
/ mo
For scale, compliance, and multi-region.
- HSM-backed vault · SOC 2 Type II
- 99.9% SLA · every rail adapter
- Agent-to-agent delegation
- Full reputation network access
- SSO, RBAC, dedicated support
Ready to give your agents a wallet?
We're onboarding ten design partners before public GA. If you're shipping agents that spend real money, let's talk.