Configuration¶

Cadence is configured via ~/.openclaw/cadence.json. The gateway reads this file at startup.

Full config reference¶

{
  "enabled": true,
  "vaultPath": "/path/to/obsidian/vault",

  "delivery": {
    "channel": "telegram",
    "telegramChatId": "123456789"
  },

  "pillars": [
    { "id": "tech", "name": "Technology", "keywords": ["code", "architecture"] },
    { "id": "business", "name": "Business", "keywords": ["strategy", "growth"] },
    { "id": "life", "name": "Life", "keywords": ["health", "habits"] }
  ],

  "llm": {
    "provider": "anthropic",
    "model": "claude-3-5-haiku-latest"
  },

  "extraction": {
    "publishTag": "::publish"
  },

  "digest": {
    "minToFlush": 5,
    "maxHoursBetween": 12,
    "cooldownHours": 4,
    "quietHoursStart": "22:00",
    "quietHoursEnd": "08:00"
  },

  "schedule": {
    "enabled": true,
    "nightlyDigest": "21:00",
    "morningStandup": "08:00",
    "timezone": "America/New_York"
  }
}

Top-level options¶

Key	Type	Default	Description
enabled	boolean	false	Whether Cadence starts with the gateway
vaultPath	string	—	Absolute path to the Obsidian vault to watch

Delivery¶

Controls where digests are sent.

Key	Type	Default	Description
delivery.channel	"telegram" \| "discord" \| "log"	"log"	Delivery channel
delivery.telegramChatId	string	—	Telegram chat ID (required when channel is "telegram")

Pillars¶

Content categories used by the insight extractor to classify extracted insights. Each pillar has:

Key	Type	Description
id	string	Short identifier (used in output)
name	string	Human-readable name
keywords	string[]	Optional keywords to help classification

Default pillars: tech, business, life.

LLM¶

Controls the model used for insight extraction.

Key	Type	Default	Description
llm.provider	"anthropic" \| "openai"	"anthropic"	LLM provider
llm.model	string	"claude-3-5-haiku-latest"	Model identifier

The LLM adapter uses OpenClaw's configured API keys. No additional key setup is needed if your gateway already has a working model provider.

Extraction¶

Controls how the insight extractor filters and processes content.

Key	Type	Default	Description
extraction.publishTag	string	"::publish"	Marker string that flags content for extraction

The extractor only processes files containing this marker. Files without it are ignored, even if they change in the watched vault.

Additional internal defaults (not currently exposed in config):

• Minimum content length: 50 characters
• Debounce window: 2 seconds after last edit
• Max batch size: 5 files per LLM call
• Min batch delay: 1 second between batches

Digest¶

Controls how insights are batched and when digests are flushed.

Key	Type	Default	Description
digest.minToFlush	number	5	Minimum insights before auto-flush
digest.maxHoursBetween	number	12	Maximum hours between flushes
digest.cooldownHours	number	4	Hours an insight must age before it can be flushed
digest.quietHoursStart	string	"22:00"	Start of quiet hours (no auto-flush)
digest.quietHoursEnd	string	"08:00"	End of quiet hours

Flush triggers¶

A digest flushes when any of these conditions are met:

1. Count trigger: Queued insights >= minToFlush, all past cooldown, not in quiet hours
2. Time trigger: Hours since last flush >= maxHoursBetween, at least one insight past cooldown, not in quiet hours
3. Cron trigger: A scheduled job fires — bypasses quiet hours and cooldown entirely

Quiet hours¶

During quiet hours (default 22:00-08:00), automatic flushes are suppressed. Scheduled cron jobs override this — if you set nightlyDigest: "21:00", that flush happens even though 21:00 is close to quiet hours, because cron triggers bypass all filters.

Cooldown¶

Freshly extracted insights wait cooldownHours before they can be auto-flushed. This prevents showing you insights from content you just wrote. Cron-triggered flushes ignore cooldown.

Schedule¶

Controls time-based triggers via the CronBridge source.

Key	Type	Default	Description
schedule.enabled	boolean	false	Whether scheduled jobs are active
schedule.nightlyDigest	string	—	Time for nightly digest flush (e.g., "21:00")
schedule.morningStandup	string	—	Time for morning signal (e.g., "08:00")
schedule.timezone	string	"America/New_York"	Timezone for all schedule times

Times are in 24-hour "HH:MM" format.

Storage¶

The digest queue is stored at ~/.openclaw/cadence/digest-queue.jsonl. This is an append-only JSONL file with line types: insight, dequeue, flush, clear. The file is fault-tolerant — malformed lines are skipped on read.

Environment variables¶

Variable	Description
CADENCE_DEBUG=1	Enable verbose signal logging on stderr

Legacy config¶

Older versions used cadence fields directly in the main OpenClaw config (~/.openclaw/openclaw.json) under the cadence key. The P1 pipeline uses the dedicated cadence.json file instead. Both are read at startup; cadence.json takes precedence for P1 options.