Platform Settings is the DataMind staff area for the platform itself. This page covers the Integrations tab: the Discord webhook, the Anthropic API key, the prompt tokenizer model and its re-weigh, the OpenRouter platform fee and the USD to GEL series. Only DataMind staff accounts can open the area; client settings live under Settings.
The tab is addressable: ?tab=integrations opens it directly, and an older address that used
notifications still resolves here. Without a tab, Platform Settings opens on Limits; the full
strip is Limits, Integrations, Email, Tool Keys, Billing Requisites and Staff.
The key pool behind web search and web read is documented under
Tool Keys.
A save is partial: only the fields you changed are sent, and everything else keeps its current
value. Empty webhook and key fields keep the stored value; clearing the tokenizer basis or the fee
returns that setting to its default. The tab confirms with Integration settings updated, and
reports Update failed when a save does not go through.
Notifications enabled turns Discord notifications on; it is off by default.
The address is stored encrypted and write-only: the card shows Configured (****{hint}) with its
last four characters, or Not configured, and never the full address. Entering a value replaces it,
and an empty field keeps the current one. An address without a protocol is refused with
Webhook must be a valid URL including the protocol (https:/.
A webhook that does not answer blocks nothing: notification sends are best-effort, and the work behind a skipped send keeps its result.
The key measures how much an agent prompt weighs in tokens, so a published version carries a real weight instead of the assumed default. It is also what the tokenizer card uses to load its model list.
The field is write-only, like the webhook: the card shows Configured (****{hint}) or
Not configured; entering a value replaces the stored key, and an empty field keeps it. The
placeholder reads sk-ant-..
Tokenizer basis selects the model whose tokenizer measures prompt weights. The card shows
In use: with the model that measures prompts today. Use default returns to the built-in
model (currently claude-sonnet-4-6); a stored id that is no longer in the list still shows as the
current selection with (no longer listed).
The basis must match whatever the premium plan currently serves: a different tokenizer generation counts the same prompt differently, so every measured weight, and the deduction built on it, would be wrong.
When the model list cannot load, the dropdown falls back to manual entry:
The model list could not be loaded, so type the id by hand: {reason}.
Changing the basis does not re-rate stored weights by itself.
After the save, the card warns: "Tokenizer model changed. Stored weights are per-version stamps taken with the previous tokenizer - saving a new model does not re-rate them. Re-weigh now so published versions carry counts from the model you just picked."
Run Re-weigh all prompt versions to measure the stored versions again. The version timeline, publishing and weights are on Agent prompt versions. Past requests are never re-rated: each one already recorded the weight it was charged.
The result block reads Checked {n} versions and groups the outcome:
re-weighed: measured again with the current tokenizer and saved.left alone, tokenizer unavailable: the tokenizer did not answer for these, so they kept the
weight they already had; nothing was overwritten.failed: these could not be re-measured, so they kept their previous weight; try again.Fee percent sets the platform fee used in the platform's financial view, where it is subtracted from every company row; it decides the net reported there. The card shows In use: with the percent that applies today. Leave the field empty to use the built-in default of 4 percent.
The value accepts 0 to 100 with up to 2 decimals, and 0 means the fee is waived:
OpenRouter fee must be a percent with up to 2 decimals, or empty for default,
OpenRouter fee cannot exceed 100%. While the default applies, the card adds
Using the platform default of 4%.
The USD to GEL series sets the conversion rate the platform's reports use; each day converts at that day's rate. A rate applies from its date until the next entry, and the earliest entry also covers every date before it; with no entries at all, the platform falls back to its built-in default rate. In force marks the entry the platform converts with today.
YYYY-MM-DD) and GEL per 1 USD (the
placeholder shows 2.), then save. The toast reads Rate added.Rate updated.Delete the rate effective {date}?. Dates the entry covered fall back to the previous rate in
the series, or to the platform default if none is left. A failed delete reports Delete failed.Rows save as you act on them, not through Save settings; the card counts its entries as
{n} entry/. When the series is empty it reads
No rates yet - the platform is using its built-in default. A refused add or edit names the
problem: Pick the date the rate takes effect on., A rate already exists for {date}.,
Rate must be a decimal with up to 6 fraction digits., Rate must be above zero.
DataMind staff accounts. The area sits outside the client console: clients never see it, and company roles cannot open it.
Both are write-only. The card shows the last four characters of the stored value (as
Configured (****{hint})) or Not configured; the full value is never displayed again. To replace
it, enter a new value; an empty field keeps the current one.
No. A save sends only the fields you changed, and everything else keeps its current value.
It activates once you change a field, and disables while the save runs.
Nothing else is affected. Notification sends are best-effort: a send that fails is skipped, and the work behind it keeps its result.
No. Stored weights are per-version stamps taken with the previous tokenizer, and a save does not re-rate them. Run Re-weigh all prompt versions. Past requests are never re-rated: each one already recorded the weight it was charged.
The list is a live reading made with the stored Anthropic key. When it cannot load, the field names
the reason: No Anthropic API key is configured - save one to load the model list,
Anthropic rejected the stored API key (401) - replace it to load the model list, or
Anthropic responded {status} while listing models. You can still type the model id by hand.
A different tokenizer generation counts the same prompt differently. If the basis does not match the model the premium plan serves, every measured weight, and the deduction built on it, is wrong.
Yes. Zero is a real setting and means the fee is waived; the built-in default is 4 percent. The value must be 0 to 100 with up to 2 decimals.
The entry with the greatest date that is not after that day; the earliest entry also covers every date before it. With no entries, the built-in default rate applies. In force marks the entry the platform converts with today.
They fall back to the previous rate in the series, or to the platform default if none is left.