A Pi OMP usage view: compare per session, not per week
Weekly totals reward whichever tool you used less. Session averages answer the routing question.
· 2 min read
A Pi OMP usage view is a desktop read of two local-first coding agents, and the fact that both write session files is what makes a single view possible. Each file records the model and the token counts for a turn, so a tracker can rebuild a day, a week or a project without a vendor account. The UsageAtlas guide covers the setup.
The two tools share a shape
| Element | Pi | Oh My Pi |
|---|---|---|
| Session store | Local files | Local files, OMP format |
| Per-turn fields | Model, tokens, timestamps | Same |
| Per-project | Working directory when recorded | Same |
| Vendor dashboard | Not aggregating | Not aggregating |
The last row is why local tracking is the practical route rather than a preference. There is nothing to log into.
What the history is good for
Three questions, each answered by a different slice of the record.
Which tool costs more per session? Divide tokens by session count instead of comparing weekly totals, which reward whichever tool you happened to use less.
Which project is expensive? Per-project grouping, when the working directory is present.
Was this week unusual? Compare with the previous three, not with a target.
Keeping the record alive
Because nobody else holds a copy, four habits matter.
- Back up the session directory with your dotfiles.
- Keep one machine as the source of truth for each project.
- Export aggregates, never transcripts, when sharing.
- Check daily so a runaway context is visible in a day.
That last one is less about discipline than about cost: a thread that grows unchecked is expensive on every subsequent turn.
Setting up
Use Pi or OMP first, install the tracker, and let it find the session directory. Today shows the live session and model split; History covers 7, 30 and 90 days.
If the app sees nothing, the directory is probably non-default. Point it there.
The limit question, again
Session files do not contain plan allowances. A local reader can report what you spent and cannot report what remains.
For a team, that means one person checks the provider page and the rest read the history. Neither source replaces the other, and the split is small enough to live with.
Why aggregates matter for privacy
A transcript contains your code. An aggregate contains a number. Exporting the second and never the first keeps a useful habit from becoming a data-handling problem.
FAQ
Do I need an account for this?
No. Local tracking needs no account, and nothing leaves the machine unless you enable optional sync.
Can the tracker see plan limits?
No. Session files record usage, not entitlement.
What if I use Pi on one machine and OMP on another?
The history will be split. Pick one as the source of truth or accept two partial views.
The short version
Two local-first tools, one shared shape, no vendor dashboard. Read the files, protect the folder, and compare per session.