Analytics and telemetry
Two views of your usage: a command on your Mac, and the console on the web.
On your Mac
optiflume analytics # all history
optiflume analytics --days 7 # last week
optiflume analytics --json # machine-readable
optiflume analytics --savings # just the savingReads the local routing log and prints the actions taken, recommendations, models used, and the estimated saving on suggestions. It counts from the local log, so its totals can differ from the console’s. Works with the network unplugged, on every plan.
What “saved” means
For each prompt OptiFlume switched: what it would have cost on the model you had selected, minus what it cost on the model that ran, at published per-token prices. It is counted only when the token counts were real (Cursor’s own usage data, or a real reading carried within the same session) and the output estimate was measured; decisions costed from a guess add $0. Claude Code and Codex do not report token counts to OptiFlume, so their switches are counted but add $0 to the saving for now. It is a counterfactual, so it is always shown as “about”. The same number is the running total shown during the trial.
On the web
The console (Pro and the trial) shows the last 30 days across every Mac on your account: Overview has the totals, Usage the detail, and Macs the activity of each Mac. To make it possible, each Mac sends a rollup of the day’s counts, at most every six hours. It is on by default so the console works the moment you sign in.
Exactly what the rollup contains
| Field | Meaning |
|---|---|
day | The calendar date of the rollup (UTC). |
prompts | Number of prompts sent to a model. |
by_tier | Prompt counts grouped by how demanding OptiFlume judged them. |
by_model | Counts per model that actually ran, by model id. |
switches | Counts of switches made automatically (accepted), models suggested without switching, prompts sent your own way with ~ or ! (bypassed), and switches stopped or kept from a switch notification. |
saved_usd | Estimated spend avoided, from published per-token prices and the model you had selected. |
saved_seconds | Reserved for time saved; always 0 today. |
compactions | Number of context compactions. Its tokens_avoided field is reserved and always 0 today. |
guard | Counts of loops caught, secrets redacted, risky commands warned or blocked. |
version | OptiFlume and Cursor version. |
device | A random id for this Mac, made once and never derived from the hardware, so several Macs add up in the console. |
trial | During the trial only: the trial meter (spend avoided, prompts, the trial's budget if any, whether it is used up, days left). |
Not in it, ever: prompt text, file names or paths, code, repository names, conversation content. The rollup is sent with your Mac’s signed license, so nobody else can post numbers into your account.
Turn it off
optiflume telemetry off # stops the rollup; local analytics keep working
optiflume telemetry on
optiflume telemetry statusThere is also a switch in console settings, for all your Macs at once; it works with every version of OptiFlume. Turning it off there stops storing rollups right away, and each Mac turns its own switch off the next time it checks in; turning it back on there turns them back on. You can also delete stored rollups there. A Mac you turned off with optiflume telemetry off stays off until you turn it on again on that Mac.