Does it read chats automatically?
No. The hosted PWA only reads files or folders you explicitly select.
FAQ
Straight answers about AI Stats.
The product favors honest limitations over imaginary automation.
Are raw chats uploaded?
No raw prompts, responses, source files, attachments, or full transcripts are uploaded in stats-only mode.
Can it track exact AI work time?
Only when the source exposes enough timing data. Otherwise values are derived, estimated, or unavailable.
Is it free?
Yes. Every available feature is included without a billing tier or product quota.
Imports and file handling
How do I get my data in? You export it from the provider, or point AI Stats at the log files a coding agent already wrote on your machine, and select those files in the browser. ChatGPT, Claude, and Gemini publish account exports. Claude Code, Codex, Gemini CLI, and Cursor write local session logs or telemetry that you can select directly.
Do I have to unzip anything? No. Archives are expanded in the browser, including nested ones, up to eight levels deep. A single selection can contain up to 1,000 files or 512 MB, expanding to at most 10,000 files or 1 GB, with any one file capped at 256 MB. Those ceilings exist to stop a malformed or hostile archive from exhausting your browser, not to ration usage.
What if my export is a format you do not recognise? The preview says so before anything is sent. Unknown shapes produce warnings rather than invented numbers, and an unrecognised Cursor database version is rejected outright instead of parsed on a guess.
What happens if I import the same export twice? Sessions are matched on provider session identity and normalized content, not on the file you dropped in, so re-importing an overlapping export updates the existing sessions rather than doubling your totals.
Accuracy and what the numbers mean
Why do two providers show different depth of detail? Because they expose different data. A chat export may carry timestamps and message counts but no token usage. A coding agent with telemetry enabled may carry tool calls, subagents, model routing, durations, and token counts. AI Stats surfaces what the source actually contains.
How do I know whether a number is measured or guessed? Every metric carries a provenance label: exact, source reported, derived, estimated, or unavailable. Derived means computed from values the export supplied, estimated means modelled with stated assumptions, and unavailable means the data is missing and is displayed as missing rather than as zero.
Are the cost figures real billing amounts? No. Where token counts and a known model are both present, an approximate API-equivalent cost can be shown as an estimate. It is not a bill, it does not know your plan or subscription, and it stays unavailable when tokens or the model are unknown.
Accounts, privacy, and sharing
Can other people see my data? No. Every query is scoped to the signed-in account. The only content that becomes public is a share page you create yourself, which is an unlisted snapshot of selected fields, revocable and expirable, containing no raw conversation content.
Can I get my data out again, or delete it? Yes to both, from settings. Export produces the normalized statistics the product holds, and deletion removes the account data rather than deactivating it.
What happens if I forget my password? Recovery is manual today, because automated email verification is not built yet. You email support from the registered address and receive a short-lived, single-use recovery password. There is no public reset form.
Why are there ads on some pages? Advertising funds a product with no paid tier. Ad units appear only on public content pages after you allow advertising cookies. They are absent from the dashboard, from every signed-in screen, and from legal and support pages.
A question that is not answered here is worth asking: the documentation, the help centre, and aistats@novusstreamsolutions.com all end up in the same place.