Heartbeat token calculator

Agents that wake on a timer and call a model every time burn tokens even when nothing has changed. Edit the fleet below to see what two methods from the playbook would do: longer per-agent intervals (method 1) and a 0-token sentinel that calls the model only when state changed (method 2).

Invented example fleet · not realAn estimate, not a measurementRuns in your browserPlaybook and case study

Totals per day

Before

–

+ method 1 (intervals)

–

+ method 2 (sentinel)

–

Change

–

Price per million tokens

Your fleet

"Changed" is the share of heartbeats that actually find something new. Most find nothing.

Agent	Interval today (min)	Tokens per wake	Changed (%)	New interval (min)	Tokens a day: before / after

Price a new agent first (method 18)

Wakes every (minutes) Tokens per wake Changed (%)

How the estimate works

Before: wakes a day × tokens per wake. Method 1: the same with the new interval. Method 2: a deterministic check runs on every wake at 0 tokens and calls the model only when something changed. A longer interval makes a change more likely per wake, so the share becomes 1 − (1 − changed)^{new / old}. The same formula is in demo/heartbeat_model.py, which you can run on your own fleet file.