What people and the AI see

Each column in an exposed dataset has two settings: what the AI sees, and what people see in their own results table. The AI only ever gets what you allow. People see real values in the results table that opens above the AI’s answer, straight from your gateway to their browser, never through the AI.

Every column starts as Real value. Setup and the console mark columns that look personal, by name or by sampled values, with “Suggested: token” and a token type. A suggestion is a hint only; nothing is tokenized until you choose it and save.

What the AI sees

Setting (console)	In the setup widget	What leaves the gateway for the AI
Token	Token	A stand-in such as `EMAIL_1a2b3c4d5e6f7a8b`, never the value. The AI can still filter, count, group and join on it.
Totals only	Totals and averages only	Only `SUM`, `AVG` or `COUNT` of the column, over groups of at least 5 rows. The AI never reads one row.
Real value	Real value	The value itself.

Tokens

A token is the column’s token type followed by 16 characters. The same value always gets the same token, in every source, so EMAIL_... for one address matches across your CRM and your shop database. Setup suggests the types EMAIL, PHONE, NAME, ADDRESS, DOB, GOVID, ACCOUNT and IP. With no suggestion it offers the column’s name in capitals. You can type your own, in capital letters, digits and underscores.

Changing the token type of a column that is already a token changes its tokens from then on. Answers given before keep the old tokens, which won’t match the new ones, so Portward asks you to confirm each such change (“Change token types?”).

Tokenize every column that identifies the same person or company. A real email next to a tokenized customer name still identifies the customer.

Totals only

With Totals only, the AI may use the column only as SUM(column), AVG(column) or COUNT(column) in one plain SELECT. Groups that cover fewer rows than the smallest group size are left out of the result, and the AI is told so. The smallest group size is 5 unless your config file sets min_group_size under [gateway] (2 at least). The console shows it under Settings, Gateway, Smallest group for totals.

Two totals can still be subtracted to work out a value, such as one person’s salary from a team’s total with and without them. Use Token for anything that must never be inferred.

What people see

For a Real value or Totals only column, people see the same as the AI (“Same as the AI”). For a Token column, you choose what people see in their results table:

Setting	In the person’s results table
Real value	The value. Shown only to them, never sent to the AI.
Masked	Dots. For values longer than 8 characters, the last 4 show.
Click to show	Dots and a button. Each click is recorded in the reveal log.
Hidden	The word Hidden.

People can also type a name or email into Find in their results to keep only the matching rows. The AI never sees what they typed, and each row Find matches is recorded.

Let people send values to the AI

Send to AI (in the setup widget, People may send to Claude or ChatGPT) lets people send the real values of that token column from one answer to the AI. Their results table then shows Send real values to Claude (or ChatGPT). Only the columns you allow are sent; the rest stay tokens. Each send is recorded. It can’t be on for a Hidden column, and turning it on gives more access.

Show more to some groups

A tokenized column can show more to some groups than to everyone else, such as salary Masked for everyone and Real value for hr-managers. In the console, open Datasets, choose the source and select By group under the column’s People see:

Select Add a group, type the group and pick what it sees.
Select Save.

A group view can only show more than everyone sees; to show less, change what everyone sees. A person in several groups sees the most open view of their groups. By group appears once the column is saved as a Token. Showing more to a group asks you to sign in again.

Group views change only what people see in their own results. The AI still sees the token.

Set the rules

Console: Datasets, choose the Source, change the columns, then Save in the bar at the bottom. The bar counts unsaved changes; Discard drops them. Find a column appears when a source has 12 columns or more.
Setup widget: ask the admin connector “Set up a connection in Portward”, open the source, change the columns under Access rules and select Save access rules.
Chat: ask the admin connector, such as “Make email in sales.customers a token and show it masked to people”. It calls propose_access and shows a review card.

Changes that give more access, such as a token column becoming Real value, or people seeing more, wait for a fresh sign-in. Changes that give less are saved at once. Either way they apply to the next answer.

Columns your config file tokenizes stay tokens; setup refuses to make them Real value.

Approvals and safety: how changes that give more access are checked.