Dataset access

Two settings decide whether someone can use a dataset. The dataset has to be exposed, and if it is limited to groups, the person has to be in one of them. Both are checked on every request, and a change applies to the next question.

Expose a dataset

A new source exposes nothing. In the setup widget (ask the admin connector “Set up a connection in Portward”) or on the console’s Datasets page, each table has a switch:

• In the setup widget’s Access rules, tick each table to expose and select Save access rules.
• In the console, open Datasets, choose the Source, turn on Exposed to the AI and to people for the table, and select Save.

Exposing a dataset gives more access, so it waits for a fresh sign-in. Turning it off is saved at once. A table the source’s login can write to shows “The account can write here: can’t be exposed” and stays off. Grant that login read-only access in the database, then Re-check the source on the Sources page.

On a source connected in setup, tables added to the database later stay hidden until you expose them. If someone adds a column to a dataset after you saved its rules, the whole dataset goes offline until an admin sets a rule for the new column. The source’s status names the columns, and the console’s Overview lists it under Needs attention.

Limit a dataset to groups

An exposed dataset with no groups is open to every data user. To limit it:

1. In the console, open Access, then Datasets by group.
2. Select Change beside the dataset.
3. Type the groups, separated by commas, exactly as your directory sends them. The field suggests groups it already knows. Leave it empty for every data user.
4. Select Save.

The Who may use it column then lists the groups, or “Every data user”. A dataset can be limited to up to 50 groups.

Whether a change gives more access depends on where it starts:

Change	Saved
Limit an open dataset to groups	At once (less access)
Remove a group from a limited dataset	At once
Add a group to a limited dataset	After a fresh sign-in
Clear all groups, opening the dataset to every data user	After a fresh sign-in

Limiting datasets to groups is done in the console. The admin connector’s propose_access changes column rules and exposure, not dataset groups.

What someone outside the groups sees

Nothing. A dataset limited to groups the person isn’t in doesn’t appear in list_datasets, a query that names it is refused with “is not available”, and links to it from other datasets are left out. If you limit a dataset after someone got an answer from it, their Click to show and Send buttons on that earlier answer stop working too.

New datasets and new groups follow the same rule. A new data user, or a new user group, gets every exposed dataset that has no groups, and none of the limited ones. If your company keeps most data private to teams, limit each dataset as you expose it.

Preview who sees what

Access, then Preview, shows what someone in given groups gets on the data connector:

1. Type the groups in Groups to preview, separated by commas. It starts with your own groups.
2. Select Preview.

It lists each dataset those groups may use and, for each column, what the AI sees and what people see, including any group views. “These groups can’t use any dataset yet” means none is open to them. The preview shows rules only, never data.

Disconnect a source

On the Sources page, Disconnect removes a source added from setup or the console. Type its name to confirm. The gateway forgets its settings, rules, context and dataset groups, and the AI loses access at once. Answers already given keep their records. A source defined in the config file is removed by editing the config file.

From chat

Ask the admin connector in plain words, such as “Expose the invoices table in sales” or “Hide sales.payments”. It calls propose_access, which checks the names against the real tables and shows a review card. See Approvals and safety.

Next

What people and the AI see: set each column’s rules.