XBPP.ORG
    Open Standard

    THE PERMISSION STANDARD FOR AGENT TRANSACTIONS

    Agents can spend money. xBPP lets them prove they should. Define spending policies as data - every transaction is evaluated before execution.

    Get StartedRead Whitepaper
    agent.ts
    import { evaluate, balanced } from '@vanar/xbpp' // On-chain transferconst verdict = evaluate(  { amount: 500, currency: 'USDC', recipient: '0x1a2b...3c4d' },  balanced) // verdict.decision  → 'ESCALATE'// verdict.reasons   → ['ABOVE_ESCALATION_THRESHOLD']// verdict.message   → 'Amount exceeds escalation threshold'
    Installnpm install @vanar/xbpp
    What is xBPP

    A PROGRAMMABLE CFO FOR AI AGENTS

    xBPP (Execution Boundary Permission Protocol) provides a verifiable way for autonomous AI agents to prove they're operating within defined spending boundaries. It answers one question: "Should this agent be allowed to spend this money?"

    The protocol separates what to check (the spec) from how to enforce (the SDK). Policies are declarative JSON - portable across any implementation, any chain, any runtime.

    Three decisions, not two

    ALLOW, BLOCK, or ESCALATE. The third option is what makes autonomous agents safe - they can ask for help instead of guessing.

    Policy is data, not code

    Declarative JSON policies that any implementation can evaluate. No vendor lock-in, no proprietary runtimes.

    Zero dependencies

    The reference SDK is 600 lines with zero runtime dependencies. Embed it anywhere - browsers, servers, edge functions.

    Start Building

    EVERYTHING YOU NEED

    SPECIFICATION

    1,760-line RFC-style spec covering evaluation phases, threat model, and wire format.

    Read the Spec

    SDK

    TypeScript reference implementation with 12 policy checks, 3 presets, and x402 integration.

    View SDK Docs

    PLAYGROUND

    Interactive spec simulator - build policies, run scenarios, and see verdicts in real time.

    Try the Playground
    How it works

    THREE DECISIONS

    ALLOW

    Transaction is within all policy bounds. Proceed automatically - no human needed.

    BLOCK

    Transaction violates a hard limit. Stop immediately and return reason codes.

    ESCALATE

    Transaction is in a grey zone. Pause and ask a human before proceeding.