What is recorded about every request
Every request records the model, the tokens, the cache hits, the cost and the timing, on every account and whatever your training settings are. What that record holds, who can see it, and what it never holds.
Every request keeps a small record of the request itself: which model answered, what it cost, and how long it took. It records the request, never the content.
This happens on every account, and it is separate from the settings that decide whether your session content may be used to improve the models.
What the record holds
- The model that answered, and the effort it was asked for.
- Tokens in and out, and how many of the input tokens came from the cache rather than being sent again.
- What the request cost, and what caching saved against it.
- How long it took to start answering and to finish.
- When it happened, and which surface it came from: chat, the CLI, the IDE, or the API.
What it does not hold is anything you or the model said. No prompts, no replies, no files, no memory.
Why it is recorded
It is how your allowance is metered and how your usage page can show you where the week went. Without it there is no way to count a week of work, no way to show you which model is costing you the most, and no way to bill anything fairly.
It is not the training setting
Two settings in Privacy controls decide whether session content may be used to improve the models. Turning them off does not turn this off, because this is the record your usage is measured from, not a copy of your work.
So the honest split is: the request record is always there and holds no content, and the content settings are yours to decide. See Is your data used to train models?.
What an organization can see
An organization's owner, administrators, billing role and auditors can see the request record aggregated for their members: totals, and the same numbers broken down by member, by model, by surface, and over time. The page shows four figures, Requests, Cost, Tokens and Saved by caching, and carries its own summary line: "Counts and cost only, never the contents of anyone's work."
There is no per-request detail in that view. Content is only readable in the conversation archive, which needs content logging set to full and a role that includes reading it. See Read and export the conversation archive.
Related articles
- Usage and analyticsThe usage page in your Dropstone dashboard shows how much of your weekly allowance you have used, by day, by surface, and by model, plus how much prompt caching has saved you.
- Is your data used to train models?On an organization account, never. On a personal account, only if you leave the setting on. What is collected, what is never collected, and what deletion can and cannot undo.
- How an organization manages its dataThe three settings an owner controls for an organization, what each one stores, who can read it, and the limits of the strongest setting. Content logging, product telemetry, and retention, in one place.
- Read and export the conversation archiveWho can read an organization's recorded conversations, how to search and filter them, what the archive never holds, and how to export it as JSONL.
- Export and delete your dataHow to get a copy of everything Dropstone holds for you, and how to remove it. Conversation export, memory, project files, CLI history, and account deletion, in one place.
- How usage and limits workEvery Dropstone plan comes with a weekly usage allowance shared across chat, the CLI, and the SDK. How it is measured, when it resets, what happens when you reach it, and how to make it go further.