Skip to main content
The Agent Kit lets you ship AI agents that can hold a wallet, spend USDC, and sign Solana transactions — without ever being able to spend money they were not authorized to spend.

​
Core guarantee

Budget enforcement that cannot be bypassed. If the SDK says “no budget”, the payment does not happen. Enforcement is server-side and atomic — three independent budget layers plus a kill switch.

​
What you get

  • Programmable agent lifecycle — create, fund, pause, resume, rotate, and kill an agent
  • Per-agent USDC wallet — an isolated wallet workspace on Solana, signed by passkey at provisioning time
  • 3-layer budget guard — infrastructure spend, token usage, and on-chain payments — enforced independently
  • Kill switch — global stop that fails-safe (blocks payments, lets infra/tokens fail-open)
  • x402-pay — pay HTTP 402 protected endpoints with USDC, with approval flow for amounts above threshold
  • Event batching — llm_call, api_call, payment telemetry batched and flushed every 10s

​
Authentication

Two modes coexist on /v1/agents/*:
CallerHeaderUsed by
Integrator (you)Authorization: Bearer <jwt> or ApiKey <key>Dashboard, your backend
Agent (the SDK)X-Agent-Auth: <agentId>:<HMAC-SHA256(method+path+body, secret)>@relayerfi/agent-sdk at runtime
The HMAC header is built inside HttpClient.buildAuthHeaders(method, path, body) — agents never pass the plain agentId:secret outside that boundary.

​
SDK

The agent runtime is @relayerfi/agent-sdk — RelayerSDK facade, HttpClient, BudgetGuard, KillSwitch, EventBatcher, X402Handler, ApprovalHandler, x402fetch, and LLM wrappers for Anthropic / OpenAI / Google. See Flow Guide for the create-agent → fund → spend lifecycle, and Endpoints for the API surface.