AgentGate

The interoperability SDK for AI agent transactions. One integration, every agentic payment rail.

---

Why

AI agents are starting to spend money — booking flights, purchasing supplies, managing subscriptions — but every payment protocol (Stripe ACP, x402, Mastercard Agent Pay, Google A2A) has its own integration, its own auth flow, and its own merchant format. If you're building an agent that transacts, you're writing four integrations instead of one.

AgentGate is the abstraction layer. Register an agent, set spending policies, and call gate.transact(). The SDK auto-routes to the cheapest available protocol, enforces your policies before any money moves, and builds a cryptographic audit trail of every action. Think of it as Plaid for AI agents — except instead of connecting bank accounts, you're connecting payment rails.

Every transaction updates a trust score. Merchants query that score to decide whether to accept an agent. Good agents build reputation. Bad agents get blocked. The trust layer is what turns a convenience library into infrastructure.

Quickstart

npm install @agentgate/sdk @agentgate/core

import { AgentGate } from '@agentgate/sdk';

const gate = new AgentGate({ apiKey: 'test', environment: 'sandbox' });

const agent = await gate.identity.register({
  name: 'shopper-bot',
  capabilities: ['purchase'],
  policies: { maxTransactionAmount: 100, dailySpendLimit: 500 },
});

const result = await gate.transact({
  agentId: agent.id,
  intent: 'purchase',
  preferredProtocol: 'auto',
  item: { description: 'Cold brew', amount: 5.99, currency: 'USD', merchantUrl: 'https://cafe.example.com' },
});

console.log(result.status); // 'completed'

No API keys, no merchant setup, no config. Sandbox mode uses a built-in mock adapter — you get a working transaction in 30 seconds.

Architecture

graph TB
    Agent["🤖 Your AI Agent<br/>LangChain / CrewAI / AutoGen / custom"]

    Agent -->|"gate.transact()"| SDK

    subgraph SDK["@agentgate/sdk"]
        Identity["🪪 Identity<br/>register · verify · get"]
        Policy["🛡️ Policy Engine<br/>enforce limits before<br/>any $ moves"]
        Router["🔀 Transaction Router<br/>auto-select best<br/>protocol by fee"]
        Trust["⭐ Trust Scoring<br/>0–100 score"]
        Audit["📋 Audit Log<br/>SHA-256 hash chain"]
    end

    subgraph Adapters["Protocol Adapters"]
        Stripe["💳 Stripe ACP"]
        X402["🪙 x402"]
        MC["🔶 MC Agent Pay"]
        Mock["🧪 Mock"]
    end

    Router --> Adapters
    SDK --> Gateway["⚡ AgentGate Gateway<br/>Hono + SQLite · localhost:3100"]
    Gateway --> Merchants["🏪 Merchants / Services / Other Agents"]

    style Agent fill:#1a1a2e,stroke:#a78bfa,stroke-width:2px,color:#e2e8f0
    style SDK fill:#16213e,stroke:#a78bfa,stroke-width:1px,color:#e2e8f0
    style Adapters fill:#1a1a2e,stroke:#635bff,stroke-width:1px,color:#e2e8f0
    style Gateway fill:#1a1a2e,stroke:#22c55e,stroke-width:2px,color:#e2e8f0
    style Merchants fill:#1a1a2e,stroke:#64748b,stroke-width:1px,color:#e2e8f0
    style Stripe fill:#635bff22,stroke:#635bff,color:#e2e8f0
    style X402 fill:#0052ff22,stroke:#0052ff,color:#e2e8f0
    style MC fill:#ff5f0022,stroke:#ff5f00,color:#e2e8f0
    style Mock fill:#64748b22,stroke:#64748b,color:#e2e8f0
    style Identity fill:#a78bfa15,stroke:#a78bfa,color:#e2e8f0
    style Policy fill:#a78bfa15,stroke:#a78bfa,color:#e2e8f0
    style Router fill:#a78bfa15,stroke:#a78bfa,color:#e2e8f0
    style Trust fill:#a78bfa15,stroke:#a78bfa,color:#e2e8f0
    style Audit fill:#a78bfa15,stroke:#a78bfa,color:#e2e8f0

Loading

Protocols

Protocol	Status	Description
Stripe ACP	Active	Full Agentic Commerce Protocol — checkout sessions, shared payment tokens, HMAC signing
x402	Coming soon	HTTP 402-based micropayments with stablecoin settlement
Mastercard Agent Pay	Coming soon	Card network payments through Mastercard's agent infrastructure
Google A2A	Coming soon	Agent-to-Agent protocol for multi-agent transaction orchestration
Mock	Active	Built-in sandbox adapter — works offline, no keys needed

Adding a protocol is one file. Implement the ProtocolAdapter interface (isAvailable, supportsIntent, estimateFee, execute, verify) and call gate.registerAdapter().

Packages

Package	Description
@agentgate/core	Shared types, protocol definitions, constants
@agentgate/sdk	Client SDK — identity, transactions, trust, policy engine
apps/gateway	Hono REST API with SQLite, auth, rate limiting, audit logging
apps/dashboard	Real-time Next.js dashboard — agents, transactions, protocols
apps/docs	Fumadocs documentation site
apps/sandbox	Demo environment with mock merchants and agent scenarios

Installation

# Clone and install
git clone https://github.com/jacobgadek/agentgate.git
cd agentgate
pnpm install
pnpm build

# Start the gateway
pnpm --filter @agentgate/gateway dev        # → localhost:3100

# Start the dashboard
pnpm --filter @agentgate/dashboard dev      # → localhost:3300

# Start the docs
pnpm --filter @agentgate/docs dev           # → localhost:3200

# Run the sandbox demo
pnpm --filter @agentgate/sandbox dev

# Run examples
pnpm --filter @agentgate/example-simple-purchase start
pnpm --filter @agentgate/example-langchain-shopping-agent start
pnpm --filter @agentgate/example-crewai-booking-agent start

Core Concepts

Identity — Every agent gets a unique ID, owner, capabilities list, and spending policies. Register once, transact everywhere.

Policies — maxTransactionAmount, dailySpendLimit, allowedCategories, allowedMerchants, blockedMerchants, requireHumanApproval. Enforced before any money moves. No exceptions.

Trust — Agents start at 50/100. Successful transactions increase trust (+2). Failures decrease it (-5). Merchants use trust scores to gate access: new (0-29) → established (30-59) → trusted (60-84) → verified (85-100).

Routing — Set preferredProtocol: 'auto' and the SDK picks the cheapest available protocol for each transaction. Or pin a specific protocol when you need to.

Audit — Every action (registration, transaction, approval, denial) is logged with a SHA-256 hash chain. Each entry references the previous hash. Tamper-evident by design.

Examples

Example	What it shows
simple-purchase	Complete transaction in 17 lines
langchain-shopping-agent	LangChain DynamicTool integration — search, check trust, buy
crewai-booking-agent	Multi-agent crew with least-privilege policies

Gateway API

All endpoints require Authorization: Bearer <api_key> except health checks.

Method	Endpoint	Description
GET	/v1/health	Health check
GET	/v1/protocols	List supported protocols
GET	/v1/identity	List all agents
POST	/v1/identity/register	Register a new agent
GET	/v1/identity/:agentId	Get agent details
POST	/v1/identity/verify	Verify agent identity
GET	/v1/transact	List all transactions
POST	/v1/transact	Execute a transaction
GET	/v1/transact/:txnId	Get transaction details
POST	/v1/transact/:txnId/approve	Approve/deny pending transaction
GET	/v1/trust/:agentId	Get trust score
GET	/v1/trust/:agentId/report	Full trust report with audit trail

License

MIT