API Rate-Limit, LLM Token-Budget & Circuit-Breaker Blueprint Guide

Across ~57,600 daily burst calls. Dual RPM (Requests/Min) and TPM (Tokens/Min) limits — enforce per-execution token caps, prompt caching, and circuit-breaker spend ceilings.

Estimated monthly overage / token protection: $890/mo in prevented runaway loop token burn.

Included in Vault #03, #09, #31 & #49 (LLM Token-Budget Guard + Model Fallback) inside the $79 StackROI 50-Blueprint Vault.

Inspects Retry-After headers, enforces inter-item pacing for 120 RPM, and checks token budget ceilings before calling the downstream API:

// n8n Code Node: Adaptive Rate-Limit Pacer & HTTP 429 Retry-After Guard (openai-anthropic)
const items = $input.all();
const targetRPM = 120;
const minDelayMs = Math.max(50, Math.ceil(60000 / targetRPM));
const maxTokensPerItem = 4096;

return items.map((item, idx) => {
  const headers = item.json.headers || {};
  const retryAfterSec = Number(headers['retry-after'] || headers['x-ratelimit-reset-requests'] || 0);
  const pauseMs = retryAfterSec > 0 ? (retryAfterSec * 1000) : (idx * minDelayMs);
  const promptText = String(item.json.prompt || item.json.body || '');
  const estimatedTokens = Math.ceil(promptText.length / 4);

  return {
    json: {
      ...item.json,
      _rate_limit_guard: {
        api_profile: 'openai-anthropic',
        batch_index: idx,
        scheduled_delay_ms: pauseMs,
        estimated_input_tokens: estimatedTokens,
        within_token_budget: estimatedTokens <= maxTokensPerItem,
        idempotency_key: item.json.event_id || ('sr_' + Date.now() + '_' + String(idx))
      }
    }
  };
});

Atomically tracks daily API request & token spend and flips circuit_state = 'OPEN' if an infinite webhook loop exceeds your daily dollar ceiling:

-- PostgreSQL Daily Token-Budget & API Circuit-Breaker Ledger (openai-anthropic)
CREATE TABLE IF NOT EXISTS stackroi_api_budget_guard (
  usage_date DATE NOT NULL DEFAULT CURRENT_DATE,
  api_profile TEXT NOT NULL DEFAULT 'openai-anthropic',
  requests_count BIGINT NOT NULL DEFAULT 0,
  tokens_used BIGINT NOT NULL DEFAULT 0,
  estimated_spend_cents BIGINT NOT NULL DEFAULT 0,
  daily_limit_cents BIGINT NOT NULL DEFAULT 2500, -- $25.00/day hard ceiling
  circuit_state TEXT NOT NULL DEFAULT 'CLOSED',   -- CLOSED (allow) | OPEN (trip)
  updated_at TIMESTAMPTZ NOT NULL DEFAULT NOW(),
  PRIMARY KEY (usage_date, api_profile)
);

-- Atomic Pre-Execution Check & Increment (Trips Circuit if Daily Spend Ceiling Reached)
INSERT INTO stackroi_api_budget_guard (usage_date, api_profile, requests_count, tokens_used, estimated_spend_cents)
VALUES (CURRENT_DATE, 'openai-anthropic', 1, $1, $2)
ON CONFLICT (usage_date, api_profile) DO UPDATE
SET requests_count = stackroi_api_budget_guard.requests_count + 1,
    tokens_used = stackroi_api_budget_guard.tokens_used + EXCLUDED.tokens_used,
    estimated_spend_cents = stackroi_api_budget_guard.estimated_spend_cents + EXCLUDED.estimated_spend_cents,
    circuit_state = CASE
      WHEN (stackroi_api_budget_guard.estimated_spend_cents + EXCLUDED.estimated_spend_cents) >= stackroi_api_budget_guard.daily_limit_cents
      THEN 'OPEN'
      ELSE 'CLOSED'
    END,
    updated_at = NOW()
RETURNING circuit_state, requests_count, tokens_used, estimated_spend_cents, daily_limit_cents;