Analysis should be reversible
Most data tooling is built around getting to an answer. VeraTracer is built around the part that actually takes the time — getting there wrong first, and needing to go back.
The problem we started with
Real analysis is not a pipeline you write once. It is a sequence of attempts: load something, narrow it, look at it, realise the filter was wrong, change it, run the whole thing again. The expensive part is rarely the thinking — it is waiting for work you have already done to be recomputed because there was nowhere to go back to.
Version control solved this for code by making history cheap and branching normal. Data work never got the same treatment. You either keep a graveyard of final_v3_actual.csv files, or you re-run everything and hope.
What VeraTracer does differently
VeraTracer records the state of your workflow at every step, not just the end. Those checkpoints are derived from the script itself, so there is no bookkeeping to maintain — name a value and you have a point you can return to.
That gives you three things that are otherwise expensive:
- Re-run a range, not the file. Change your last step and only that step recomputes.
- Branch without copying. Try a second approach from any checkpoint and compare both results side by side.
- Rewind instead of replaying. Returning to an earlier checkpoint restores recorded state, so it costs the same whether the step took 20 milliseconds or 20 minutes.
Read Checkpoints for how this works in practice.
Where the platform is today
Being specific about this matters more than sounding finished:
| Piece | Status |
|---|---|
| VeraTracer Desktop | Launching December 13, 2026 on macOS. Windows and Linux follow. |
| VeraScript | Working today, with four statements — see the reference. |
| Cloud runs | In development. Not yet available. |
| Connectors | In development. Local CSV works today. |
How it is built
VeraTracer Desktop is a native application — a Rust core with a web-technology interface, shipped through Tauri rather than a browser tab pretending to be an app. It opens real folders on your disk, reads and writes real files, and runs workflows entirely on your own machine. Nothing is uploaded when you run locally.
We publish what changes in the changelog, including the things we deliberately have not built yet.
Get in touch
Questions, or something that should work and does not? The contact page has the ways to reach us.