#Heartbeat Token Cost Analysis

Deep analysis of OpenClaw's cron heartbeat mechanism and its token cost implications.

#Source

src/cron/service/timer.ts (877 lines)

#What is a Heartbeat?

A heartbeat is a full context sent to the LLM every 600 seconds (10 minutes) expecting a simple "HEARTBEAT_OK" response. This keeps long-running cron jobs alive and responsive.

export const DEFAULT_JOB_TIMEOUT_MS = 10 * 60_000; // 10 minutes

#Cost Model

#Token Consumption Per Call

Pricing assumptions (Claude Opus 4.6):

• Input: $3/M tokens
• Output: $15/M tokens

#Daily Cost Projections

At 600-second (10-minute) intervals = 144 heartbeats/day

#Monthly Projections (30 days)

#Error Backoff Strategy

When heartbeat errors occur, OpenClaw applies exponential backoff:

// From src/cron/service/timer.ts
const ERROR_BACKOFF_SCHEDULE_MS = [
  30_000,      // 1st error  →  30 seconds
  60_000,      // 2nd error  →   1 minute
  5 * 60_000,  // 3rd error  →   5 minutes
  15 * 60_000, // 4th error  →  15 minutes
  60 * 60_000, // 5th+ error →  60 minutes
];

function errorBackoffMs(consecutiveErrors: number): number {
  const idx = Math.min(consecutiveErrors - 1, ERROR_BACKOFF_SCHEDULE_MS.length - 1);
  return ERROR_BACKOFF_SCHEDULE_MS[Math.max(0, idx)];
}

MAX_SCHEDULE_ERRORS = 3 before job is disabled.

#Timeout Configuration

export const DEFAULT_JOB_TIMEOUT_MS = 10 * 60_000; // 10 minutes

function resolveCronJobTimeoutMs(job: CronJob): number | undefined {
  const configuredTimeoutMs =
    job.payload.kind === "agentTurn" && typeof job.payload.timeoutSeconds === "number"
      ? Math.floor(job.payload.timeoutSeconds * 1_000)
      : undefined;

  if (configuredTimeoutMs === undefined) {
    return DEFAULT_JOB_TIMEOUT_MS;
  }

  return configuredTimeoutMs <= 0 ? undefined : configuredTimeoutMs;
}

Timeout per job execution: 10 minutes (600,000ms)

#Optimization Recommendations

#1. Increase Heartbeat Interval

Current: 600 seconds (10 minutes) Proposed: 1800-3600 seconds (30-60 minutes)

Savings: 3-6x reduction in daily costs

#2. Use Lightweight Health Check

Replace full context with minimal status check:

• Send only: job metadata + last run status
• Expected response: simple "OK" or status enum
• Token savings: 95%+ reduction

Estimated cost with lightweight check:

• Input: ~500 tokens (metadata only)
• Output: ~100 tokens
• Cost/call: ~$0.003
• Daily cost: ~$0.43 (vs. $86.40)

#3. Conditional Heartbeat

Only send heartbeat when:

• Job has pending work
• Last run was > 1 hour ago
• User explicitly configured heartbeat: true

Savings: 50-80% reduction depending on job activity patterns

#4. Context Pruning for Heartbeats

Apply aggressive compaction before heartbeat:

• Keep only: last 5-10 messages
• Summary of older history
• Token savings: 60-80% reduction

#Trade-offs

#Recommended Hybrid Approach

1. Increase interval to 30 minutes (immediate 3x savings, minimal risk)
2. Add lightweight check between full heartbeats (check every 5 min, full heartbeat every 30 min)
3. Apply context pruning to full heartbeats (keep last 10 messages + summary)

Expected savings: 80-90% cost reduction with acceptable trade-offs

#Cross-References

• Cron System
• Context Budget Allocation