Save tokens and costs

[dim hawk]

How to reduce token costs?
by /reset(on telegram? )doing/new eveytime?

its running in background and suing up my tokens

@grand glen

?

[grand glen]

Yes, /new or /reset gives you a fresh session, so it usually lowers token use versus carrying a long chat forever.

If tokens are being burned “in the background”, that is usually one of these:

1. Heartbeat

   • disable it: agents.defaults.heartbeat.every: "0m"
   • or make it cheaper with:
     • lightContext: true
     • isolatedSession: true
     • a cheaper heartbeat.model

2. Long sessions

   • use /new or /reset
   • or /compact if you want to keep some context but shrink it

3. Expensive model

   • switch the current session with /model

• or set a cheaper default model in config

4. Check what is actually costing tokens
   • /status
   • /usage cost

Docs:

• Sessions: https://docs.openclaw.ai/concepts/session
• Token use: https://docs.openclaw.ai/reference/token-use
• Heartbeat: https://docs.openclaw.ai/gateway/heartbeat
• Models: https://docs.openclaw.ai/concepts/models

If you want, paste your heartbeat and model config and I’ll tell you the cheapest setup.

[spark sequoia]

Im using Minimax for easy stuff it is very cheap

[grand glen]

Yep, that’s a good strategy.

Using a cheap model for routine chat helps a lot, but the sneaky part is background turns:

• heartbeat
• cron jobs
• long old sessions with huge context

So I’d do both:

1. keep Minimax for cheap/easy stuff
2. make sure heartbeat/background also uses that same cheap model
3. use /new or /reset sometimes, because even a cheap model gets more expensive with a huge session history

If you want the biggest win, check these:

• /status to see session/context size
• /usage cost to see where spend is going
• heartbeat config: https://docs.openclaw.ai/gateway/heartbeat