Base Agent API Documentation

Select an Endpoint

Choose an endpoint from the left menu to view documentation and test live responses.

Architectural & Verification Specifications

1. 4-Step Agent Policy-Pipeline Architecture

Static bytecode analysis produces capability signals, not autonomous enforcement. Consuming AI agents and bots implement our standardized 4-step policy model:

                            Step 1: Observed Facts

                            Extract PUSH-aware opcodes (e.g. DELEGATECALL) and 4-byte selectors (e.g. pause()).
                        
                            Step 2: Risk Implication

                            Structural risk (e.g. implementation logic can mutate or token transfers can be halted).
                        
                            Step 3: Consumer Policy

                            Agent rule evaluation (e.g. if (hasPause && !allowPaused) return BLOCK;).
                        
                            Step 4: Execution Decision

                            PERMIT or BLOCK transaction before gas commitment.

2. RPC Trust Hierarchy & Quorum Consensus

Every bytecode analysis and data point carries cryptographic provenance and trust grading:

Trust Level	Definition & Requirement	Evidence Grade
`HIGH_TRUST_PRIMARY`	Authenticated, TLS 1.3-enforced Base RPC endpoint configured by operator (e.g. Alchemy, QuickNode) with Chain ID 8453 validation.	TRUE
`QUORUM_PUBLIC`	Consensus agreement across ≥ 2 independent public Base mainnet RPCs (Base.org, 1RPC, DRPC, Blast, MeowRPC).	TRUE
`DEGRADED_LOW_TRUST`	A lone public endpoint answered without quorum verification. Capability rating degrades to `UNVERIFIED`.	FALSE
`UNTRUSTED_INSECURE`	Insecure HTTP or invalid chain data. Fails closed with HTTP 503.	FALSE

3. Latency & Performance Benchmark Distribution

Measured on Node.js 22 runtime on AWS IAD1 directly connected to Base Mainnet:

In-Memory Disassembly:

< 15 ms

p50: 8ms | p95: 14ms

Server Warm RPC Preflight:

< 35 ms

p50: 24ms | p95: 38ms

Global Client Roundtrip:

45 – 120 ms

Edge routing dependent

Serverless Cold Start:

~120 – 180 ms

Initial container boot

4. Real-World Case Study: Autonomous DEX Swap Preflight Guard

Measured Latency: 28ms

When an autonomous agent or trading bot executes swaps on Base (e.g. Aerodrome, Slipstream, Uniswap V3), an uninstantiated proxy slot or active freeze selector can lock funds or cause silent transaction reverts. M2M Sentinel acts as an inline deterministic circuit-breaker before gas commitment:

const { M2MSentinelClient } = require('m2m-sentinel-sdk');
const sentinel = new M2MSentinelClient({ apiKey: process.env.M2M_SENTINEL_API_KEY });

async function preflightSwapGuard(targetToken) {
  // 1. Fetch deterministic static capability telemetry (28ms)
  const audit = await sentinel.auditContract(targetToken);

  // 2. Reject uninstantiated / broken proxies before signing
  if (audit.proxy && audit.proxy.isProxy && !audit.proxy.implementationAddress) {
    throw new Error(`[REJECTED] Broken proxy with null implementation slot.`);
  }

  // 3. Evaluate observed freeze / pause capability flags against policy
  const capabilities = audit.observedCapabilities || [];
  if (capabilities.includes('FREEZE_SELECTOR')) {
    throw new Error(`[REJECTED] Target contract contains active freeze selector.`);
  }

  return { verified: true, trust: audit.trustLevel };
}