The all-companies request log shows every company's requests in one table. It is the internal view of Usage: DataMind staff use it to answer support questions and to trace a single request back to the upstream provider.
This is a staff view. A company account on the same screen sees only its own requests, with the backing model, the provider and the cost figures masked.
Open Usage in the console. A staff account gets the Internal panel: the Request log
section, subtitled Internal view · all companies · real backing models. A company account sees its
own Logs panel instead (Request log).
The window picker at the top starts on the current month and sets the billing window for the rows; it works as on the client Usage screen. Rows are newest first.
The view has no company scope of its own: a company is one more column and one more filter. The Company column comes first.
To pin the log to a single company:
Current company · {name}, to set the filter to that
company. Click it again to clear the filter.Company names are unique, so the filter matches one company exactly. The filter set is shared: the token tiles above the log follow it too.
The rows are unmasked: the model, the provider and the cost figures are the real ones.
These are the columns the internal view adds, or reads differently from the client log.
| Column | What it shows |
|---|---|
| Company | The company that made the request. |
| Model | The real backing model that served the request (Model (backing)). |
| Requested | The model the caller asked for (Requested model). |
| Provider | The real serving provider (Serving provider). Plan requests are not relabelled with a region label here; the client log shows EU Provider or Global Provider instead. |
| Cached | Prompt tokens served from the cache (Cached tokens (not deducted from billable tokens)). |
| Error | The upstream error message on a failed request, shortened (Error message (staff only)). |
| Error code | The error code on a failed request. |
| Version | The client build version the caller reported (Client build version); blank when it reported none. |
| Upstream | DataMind's upstream cost for the request, in USD. |
| Base | The pre-markup basis: the billed cost divided by the markup. Pay-as-you-go rows only. |
| Margin | The billed cost less the base. Pay-as-you-go rows only. |
| Stream | A tick when the response was streamed. |
| Generation ID | The upstream generation reference. Hidden by default; show it from the columns menu. |
The remaining columns read as on the client log: Time, Agent, Bucket, Region, Input, Output, Kartlos Discount, Total Tokens, Latency, Status, Finish, Billed, Cache Disc and External ID (Request log).
Search, column filters, sorting and column visibility work as on any table
(Tables overview); the filter menus offer the values present in the
window. A window or filter set with no rows reads No requests match this window.
Export data covers every row matching the current filters, not only the rows on screen, and streams from the server as CSV or Excel (Table export, copy and upload).
Every row records the upstream generation reference in Generation ID. Copy the value and look it up in the upstream provider's own records; that is how one request is traced to the end. The column is hidden by default, so show it from the columns menu first.
A failed request keeps a shortened copy of the upstream error message in Error; open the cell to read the whole kept copy. Upstream spend is reconciled against the ledger on Upstream usage and reconciliation.
DataMind staff accounts. On the same screen, a company account sees only its own requests, with the backing model, the provider and the cost figures masked (Request log).
Open the Company column's filter menu and click the Current company preset to pin the log to the company you are switched into; click it again to clear. You can also filter the Company column directly; names are unique, so the match is exact.
Requested is what the caller asked for; Model is the real backing that served the request. Compare the two when you need to know which backing answered.
Copy its Generation ID and look the value up in the upstream provider's own records. The column is hidden by default; show it from the columns menu first.
Upstream and Cache Disc fill in during reconciliation, so the newest rows can under-report until it catches up.
The Version column shows the build version behind a request; it is blank when the caller reported none. For the latest build a company has reported, open the companies list (Companies); a malformed value is ignored, so a broken stamp never shows as a version.
The Error column keeps a shortened copy of the message the upstream returned; open the cell to read the whole kept copy. The Generation ID takes you to the upstream record.
Yes. The staff view reads the whole ledger, so embeddings requests appear here; the client log leaves them out.