#it the "Read HEARTBEAT.md ... " using up all tokens?
1 messages Β· Page 1 of 1 (latest)
Yes, heartbeats do consume tokens! Each heartbeat check-in runs your configured agent model, which means API calls and costs.
However, $13 is extremely high for just heartbeats. A typical heartbeat uses maybe 1-5K tokens per check, and with reasonable settings (e.g., every 5-15 minutes), you'd be looking at cents per day, not dollars.
Possible causes:
- Frequent heartbeat intervals β If you have heartbeats running every minute or very frequently, those costs add up fast
- Heavy agent logic β If the agent is doing substantial work on each heartbeat (checking emails, calendar, weather, etc.)
- Expensive model β Using a premium model like opus-4 for heartbeats instead of a cheaper one like haiku or minimax
- Multiple agents β More than one agent running heartbeats simultaneously
- Bug/loop β Something might be stuck in a loop
To investigate:
- Check
agents.defaults.heartbeat.everyin your config - See what model your agent uses for heartbeats vs. normal responses
- Check if you have multiple agents configured with heartbeats
Quick fix: Set a cheaper model for heartbeat (like minimax/minimax-m2.1 or openai/gpt-4o-mini) and increase the interval.
Want me to point you to the relevant docs? https://docs.clawd.bot/concepts/models has model pricing comparisons. π¦
@distant depot if the context window is quite full then heartbeats consume loads of tokens.
@wheat iron how to set a cheaper model for heartbeats only though?
Yes! Heartbeats DO consume tokens - especially when the context window is full, as the full context gets sent with each heartbeat turn.
Good news: you can set a cheaper model specifically for heartbeats. Here's how:
{
"agents": {
"defaults": {
"heartbeat": {
"every": "30m",
"model": "minimax/MiniMax-M2.1", // π cheaper model here!
"target": "last"
}
}
}
}