SquidSupport
Product reference / AI

Config set details

In the app: AI → Advanced → Config Sets → click a config set

Needs the 'View AI Config' permission.

The page of one config set: its name, the stores that use it, whether it is answering customers, and a few settings that belong to the set as a whole. The prompts, specialists, tools and variables inside the set are edited under AI → Advanced settings; this page counts them and links there.

Header actions

The header shows the set's name, a "Default" badge when it is the organization's default set, and its description. "Back to AI Config Sets" returns to the list.

People with the 'Edit AI Config' permission also see:

  • "Discuss & improve in Builder": starts a Builder session for this set and opens it. The session edits a draft; saving a draft keeps the work in the session. Publishing applies it to a config set.
  • "Set as Default": makes this set the default for stores with no assignment of their own. Hidden on the set that is already the default. A teammate whose access is limited to certain stores cannot change the default, unless the stores they have access to are all the stores of the organization. A single-store set cannot be the default. Unresolved failed Builder policy checks for the newly affected stores can also block the change.
  • "Clone": asks for a name, proposed as the set's name with "(Copy)", and creates an independent copy with the same prompts, specialists, tool settings, variables and offers. No store is assigned to the copy.
  • "Delete": opens "Delete Config Set". While stores are still assigned, the dialog only says how many and asks you to reassign them first. Otherwise it deletes the set with all its prompts, specialists, tool settings and variables, which cannot be undone. A set with version control switched on must have that switched off first. The default set has no "Delete" button.

Without the edit permission the page is read-only.

Activate and stop answering

Needs the 'Edit AI Config' permission.

A banner under the header says whether the set is answering customers, and for which store.

When it is not answering yet, the banner reads "Not answering customers yet" and offers "Activate". With more than one assigned store that is not live, "Store to activate" chooses which one. Activating does four things together, for that one store:

  • it makes the newest version of every prompt in the set active;
  • it switches on every specialist in the set;
  • it assigns the set to the store;
  • it switches the store's AI agent on.

Either all four happen or none. Whenever activation finds the store’s AI switched off, it starts in draft mode: replies wait in your inbox for approval, and each one uses an AI message from your plan. A store whose agent was already answering keeps its own choice between drafts and automatic sending. The agent starts answering as soon as a customer writes in. "Activate" needs a store: one assigned below, one named by the link that opened the page, or the store you have selected in the app. With neither, the button is off and the banner asks you to assign a store first. If activation is refused, the message under the button names what to fix, for example a missing agent name.

Unresolved failed Builder policy checks for the affected stores and channels can refuse activation.

Because activating makes the newest prompt versions active for the whole set, it also changes what other stores on the same set get.

When it is answering, the banner says so, for example "Answering for … · replies wait in your inbox" or "… replies are sent to customers automatically, without review", and lists the other stores the set is live for. "Stop answering" switches the AI agent off for one store after you confirm; with several live stores, "Store to stop" chooses which. Stopping leaves the set, its prompts and its assignment as they are, so you can switch it back on. New conversations of that store then wait for your team.

Insights

The insights card shows how this configuration did recently: the AI handling rate, the counts of handled and touched conversations over the last 30 days, the most common reason for handing over, and negative QA review figures when there are any. The percentage divides handled conversations by the total handled, escalated or ignored. That can differ from the touched count printed beside it. When the summary is unavailable it says "Insights will appear once this configuration has been deployed and seen real traffic."

Details

  • "Name": the set's name, as shown in lists and pickers.
  • "Description": a note for your team on what the set is for.

"Save Changes" appears once you change either. The name must not be blank. An empty description is omitted when saving, so clearing this field does not remove the saved description. Renaming a set does not change anything for customers.

Release tags are not set here: they belong to a release and are set on the "Version control" tab of the set, which the line under the fields links to.

Return instructions

Return instructions are the wording customers get for sending a return back: shipping costs, required forms and documents. They are shared by every store that uses this config set and are kept per language.

"Add return instructions" opens a form:

  • Language: a language code such as "de". There is one set of instructions per language.
  • Instructions: what the customer must do, up to 8,000 characters. Do not repeat the return address; it is added automatically.
  • "Return form URL (optional)": a link to a return form. Only full http or https web addresses with a host are accepted, up to 1,000 characters.
  • "Proof of postage required": when on, the AI is told the customer must keep proof of postage.

When the AI explains how to send a return back, it uses the instructions in the language of its reply. If there are none in that language, it uses the ones in the store's main language. If neither language has instructions, or they cannot be read, the AI sends the return address on its own.

Each row can be edited or deleted. Deleting a language's instructions means replies in that language fall back as described. Instructions cannot be changed on a release.

Custom sign off

"Enable custom config set sign off" is off by default.

When it is on and "Custom sign off" is filled in, emails sign off with this text instead of the store's own sign-off from Settings → Email. An AI reply uses the sign-off of the config set that wrote it. Other emails from the store, including replies written by your team, use the sign-off of the config set the store uses for email. When the switch is off or the text is empty, emails use the store's sign-off.

"Save Custom Sign Off" stores the switch and the text. The preview uses the store's and agent's names when exactly one store is assigned. The sign-off cannot be changed on a release.

Resources

The "Resources" card counts what is inside the set:

  • Prompts: the number of prompts, with the number of versions and how many are active. "none active" means the set has prompts but none is switched on.
  • Specialists: the specialist agents in the set.
  • Tool overrides: tools switched on or off for this set. "none — every tool uses its default" means nothing was changed.
  • Variables: the reusable facts the prompts use.

"Edit in AI Agent settings →" opens AI → Advanced settings for this set, where all of these are edited.

A warning appears when the set has prompts but no active main prompt. It says the set is answering nobody. A store can still answer using an active main prompt saved directly for that store; specialists alone do not provide a main prompt. "Activate" above switches the prompts and specialists on together.

Assigned Stores

The list shows every store that uses this set, with badges:

  • "AI Enabled" or "Not answering yet": whether the store's AI agent is switched on;
  • "Auto-send": the store sends replies without review;
  • "Default", or a channel name: whether the store uses the set for everything or only as an override for chat (the chat widget and WhatsApp) or for email.

"Assign Store" lists the stores that do not use this set. Choosing one asks you to confirm, because the store then stops using the set it is on now and answers from this one, as its store-wide set. Assigning does not switch the store's AI on or off. If the store's AI is on, subsequent replies use this assignment where a channel override or an active AI test does not select a different configuration.

While the store's AI is switched on, assigning is refused, with a link to the place to fix it, when neither the set nor the store has an active main prompt, or when specialists referenced by the main prompt are not ready. A store whose AI is off can be assigned to such a set. A single-store set can only be assigned to its own store. If its original store was deleted, the next assigned store claims it. Unresolved failed Builder policy checks can block assignment even while the AI is off. Releases cannot be assigned directly.

If this set uses tools that need an integration the store has not connected, the store is still assigned. Those tools are switched off for that store only, a note under the store says which and why, and they switch on by themselves once the integration is connected.

"Unassign store" (the cross on a row) removes the store from this set after you confirm; each of this set’s store-wide and channel assignments is cleared. A cleared channel override falls back to the store-wide set, then the organization’s default. Other channel overrides stay in place. Unresolved failed Builder policy checks on the fallback can refuse a removal. The default set has no unassign button, because there is nothing to fall back to.