Capri Ai Agentpay

Capri AgentPay is the governed payment layer for AI agents, designed to solve the security nightmare that happens when you hand a private key to an autonomous system. In the current landscape, giving an agent access to a wallet is basically asking for trouble. One bad prompt, one hallucinated API call, and your entire balance can vanish before you even notice. Capri eliminates that risk by inserting a runtime between the agent and the signer. Think of it as a bouncer for your wallet. The agent never touches the keys. Instead, it calls a single typed surface, and when it hits an HTTP 402 Payment Required challenge, Capri steps in to evaluate your policy in a fixed order: merchant allowlist, per-transaction cap, daily budget, and an auto-approve threshold that can pause for human review. Only when every check passes does it settle, retry the original request with proof attached, and store a signed receipt you can verify offline. This product is built for teams shipping agent systems in production who need real controls around money, approval workflows, and live infrastructure. It gives operators clear review queues, audit trails, and boundaries that do not collapse back into manual busywork. Capri AgentPay is the first shipping layer in the Capri stack, and it is currently available in a Base Sepolia sandbox for testing, with a VS Code extension and six MCP tools that replace raw key access with narrow, policy-gated operations. If you are tired of being scared to let an agent near a paid API, this is your solution.

x402 Payment Handling

Capri AgentPay natively handles HTTP 402 Payment Required challenges, which is the emerging standard for machine-to-machine payments. When an agent calls a paid API or MCP tool and gets hit with a 402, Capri intercepts that response, evaluates your policy, and either approves or denies the payment automatically. This means your agent does not have to know anything about wallets, signing, or gas fees. It just makes requests, and Capri handles the financial plumbing. The result is a seamless experience where the agent focuses on its task while the payment layer stays invisible and secure.

Per-Agent Budgets and Caps

You can set granular spending limits for each individual agent, including per-transaction caps and daily budgets. This prevents a single runaway agent from draining your entire treasury. For example, you can give your research agent a $50 daily budget with a $5 per-call cap, while your data analysis agent gets $200 daily with a $20 cap. These limits are enforced at runtime before any signature is created, so even if an agent goes rogue, it can only spend what you allow. This feature gives you surgical control over costs without micromanaging every single API call.

Merchant Allowlists

Capri AgentPay lets you define explicit lists of approved merchants or API endpoints that your agents are allowed to pay. This is your first line of defense against malicious or hallucinated calls. If an agent tries to send money to a domain that is not on the allowlist, the transaction is automatically blocked before it ever reaches the signer. You can maintain separate allowlists for different agents or teams, giving you the flexibility to lock down production agents while allowing development agents to test against new services. It is a simple but incredibly effective way to prevent catastrophic mistakes.

Human-in-the-Loop Approvals

When a payment request exceeds your auto-approve threshold, Capri pauses the transaction and sends it to a human operator for review. The operator sees the exact amount, asset, chain, and recipient before making a decision, along with the policy branch that triggered the review. This gives you a safety net for high-value or unusual transactions without slowing down the vast majority of routine payments. You can set the threshold at any level, from $0 (require human approval for everything) to unlimited (auto-approve all transactions within budget), depending on your risk tolerance and operational needs.

Verifiable Receipts

Every settled payment produces a signed receipt that includes hashes, timestamps, and verification fields. These receipts are stored and can be verified offline, giving you a cryptographic proof of payment that does not rely on Capri's servers. This is critical for audit trails, compliance, and dispute resolution. If a merchant claims they were not paid, you can pull the receipt and verify it independently. The receipts also include the original request context, so you know exactly which agent made the call, to which endpoint, and for what amount. It is the kind of transparency that security teams love and regulators require.

Six MCP Tools Instead of Raw Key Access

Instead of giving your agent direct access to a private key or wallet, Capri provides six narrow MCP (Model Context Protocol) tools that cover all common payment operations. These tools include fetching balances, checking policies, requesting payments, viewing receipts, and managing allowlists. Each tool is individually policy-gated, so the agent can only do what you explicitly allow. This drastically reduces the attack surface compared to giving an agent a raw signing key. If an agent is compromised, the attacker can only use these six tools within your defined policies, not drain your entire wallet.

Autonomous Research Agents Accessing Paid APIs

Research agents often need to call paid APIs for premium data, such as financial market feeds, scientific databases, or proprietary AI models. Without Capri, you either give the agent a key and pray it does not go nuts, or you manually approve every call. With Capri, you set a daily budget and an auto-approve threshold, and the agent can autonomously access the data it needs. If a single call exceeds your threshold, a human reviews it. The result is a research agent that can operate 24/7 without burning through your budget or exposing your keys.

Automated Trading and Financial Operations

Trading bots and financial agents need to settle transactions on-chain, but giving them a private key is a security nightmare. Capri solves this by acting as a governed runtime between the agent and the blockchain. You set per-transaction caps, daily budgets, and merchant allowlists for approved DEXs or protocols. The agent can execute trades within these boundaries, and any unusual activity triggers human review. The verifiable receipts provide a clear audit trail for compliance. This lets you deploy trading agents in production without losing sleep over a catastrophic loss.

Internal Tool Orchestration with Budget Controls

Enterprises often have internal APIs that cost money per call, such as data transformation services, image generation endpoints, or specialized compute resources. Capri lets you give agents access to these tools with strict budget controls. Each team or department can have its own budget, and the merchant allowlist ensures agents only call approved internal services. The human-in-the-loop approval catches any unusual spending patterns. This turns your internal API ecosystem into a governed marketplace where agents can operate autonomously but within financial guardrails.

Customer-Facing Agent Services with Usage Limits

If you are building a product where customers interact with AI agents that make paid API calls on their behalf, Capri is essential. You can set per-customer budgets, auto-approve small transactions, and require human review for larger ones. The verifiable receipts give your customers confidence that their funds are being used correctly. The merchant allowlist prevents agents from calling unauthorized services. This allows you to offer premium agent services without the risk of a single customer's agent running up an enormous bill or sending money to a scammer.

Does the agent ever hold my private keys?

No. The agent never has access to your private keys. Capri operates as a runtime between the agent and the signer. The agent calls a typed surface, and when a payment is needed, Capri evaluates your policy. Only after every policy check passes does a signature get created. This means even if an agent is compromised or hallucinates a malicious call, your keys are safe. The worst that can happen is the agent spends up to your defined limits, which is a known and manageable risk.

What happens if a payment exceeds my daily budget?

If an agent tries to make a payment that would exceed its daily budget, Capri blocks the transaction immediately. The agent receives an error response indicating that the payment was denied due to budget limits. You can then review the agent's activity, adjust the budget if needed, or investigate why the agent is trying to spend more than expected. This hard stop prevents runaway spending and gives you time to intervene before any money leaves your wallet.

Can I use Capri AgentPay with any blockchain or payment network?

Currently, Capri AgentPay supports settlement on Base Sepolia for sandbox testing. The architecture is designed to be chain-agnostic, and support for additional chains and payment networks is on the roadmap. The x402 standard is network-agnostic, so Capri can theoretically work with any chain that supports the appropriate smart contracts. Check the documentation for the latest supported networks, as the team is actively expanding compatibility.

How does the human-in-the-loop approval work in practice?

When an agent makes a payment request that exceeds your auto-approve threshold, Capri pauses the transaction and sends a notification to the designated reviewer. The reviewer sees a clear interface showing the exact amount, asset, chain, recipient, and the policy branch that triggered the review. They can approve, deny, or modify the request. If approved, Capri proceeds with settlement and retries the original request. If denied, the agent receives an error. This whole process takes seconds and prevents high-value mistakes without slowing down routine transactions.

Pricing information is not available in the provided content. Capri AI is currently opening access in small waves to teams building agent systems. To get pricing details, you should request access through their website or book a demo. The Base Sepolia sandbox is available for testing, which suggests there may be a free tier for development, but no official pricing plans or tiers have been disclosed in the context provided.