What You Need to Know
A standard OpenClaw session loads all workspace files at startup: SOUL.md, AGENTS.md, USER.md, IDENTITY.md, TOOLS.md, and more. For interactive conversations, this full context is essential โ the assistant needs your personality, rules, and preferences to respond well. But for a heartbeat check that just reads a checklist and reports status, loading 5,000+ tokens of workspace context is wasteful.
Light bootstrap addresses this with agents.*.heartbeat.lightContext: true in your config. When enabled, heartbeat runs load only HEARTBEAT.md. The assistant gets its checklist, executes the checks, and reports results โ without consuming tokens for personality, communication style, or workflow rules that are irrelevant to a background health check.
For cron jobs, the equivalent is the --light-context flag. Add it to your cron job definition to run scheduled tasks with minimal context loading. A cron job that checks RSS feeds, monitors disk usage, or fetches weather data does not need your full agent personality โ it needs its task instructions and nothing more.
The token savings compound over time. If your HEARTBEAT.md is 200 tokens but your full workspace context is 3,000 tokens, every heartbeat run saves 2,800 input tokens. At 48 heartbeats per day (every 30 minutes), that is 134,400 tokens per day โ significant enough to change your monthly cost profile, especially on premium models.
Light bootstrap is opt-in because some heartbeat tasks genuinely benefit from full context โ for example, a heartbeat that needs to reference your communication preferences when formatting an alert. Start with light context for simple check-and-report heartbeats, then add full context back selectively for tasks that need it.