AI memory, rules and suggestions

Structurell's AI doesn't change your data on its own. What it proposes lands as a suggestion for a person to review, and decisions it should make the same way every time can be written down as rules. The pages in this article are under Account → Settings → AI, next to your AI Brain. If you haven't set up your brain yet, start with AI settings and your AI brain.

Before you start

• These pages are for subscription administrators.
• AI must be included in your subscription's package. See Package and add-ons.
• Each page has a Refresh button at the top to reload the latest information.

Update memory

Update memory is a chat with the AI about your brain. Ask planning questions, or ask about something it should know, such as a new service you offer or a phrase you never want used.

1. Open Update memory.
2. In Brain profile, keep Use default brain or pick another profile.
3. In Applies to, choose the area of Structurell your question is about, such as Catalogue, Content or Social.
4. Type your question in Message and select Send message.

The answer appears in the Reply panel. To carry on the same conversation, send another message; the Conversation ID is filled in for you. Used for and Advanced context are optional and can be left empty.

If the AI thinks your brain should change, it adds a suggestion to the Suggestion queue rather than changing anything itself. To change a memory, edit it under AI Brain → Advanced settings. See AI settings and your AI brain.

Suggestion queue

The Suggestion queue lists AI suggestions, such as ideas from Update memory and category suggestions for products. Filter the list by area, target type and status. Suggestions can be Drafted, Approved, Applied, Rejected or Failed.

Select a suggestion to see its detail, including the reasoning and confidence the AI gave. Then:

• Approve: records that you agree with the suggestion. You're asked to confirm with Approve this suggestion?
• Reject: turns the suggestion down. You can add Feedback to explain why. Your feedback is saved to the AI's memory so it's taken into account next time.
• Apply with approval: sends the change to the part of Structurell it's for: your catalogue, content or social posts. You're asked to confirm with Apply this suggestion to the target service? If the suggestion's area doesn't support this, you'll see Apply is unavailable for this service. If the change is refused, the message explains why.

Approved rules

Rules are fixed instructions the AI checks before it tries to work something out itself. For example, a rule might say that any product whose text mentions both "coffee" and "beans" is tagged "coffee". Because rules always give the same answer, they're predictable and easy to audit.

Approved rules lists every rule with its name, what it Applies to, what it's Used for and its status: Proposed, Approved or Disabled. Use the filters to narrow the list. For each rule you can:

• Edit: load it into the form beside the list.
• Approve: make it active. You're asked to confirm with Approve this deterministic rule?
• Disable: stop it being used without deleting it.
• Delete: remove it after confirming.

To create a rule, select New, then fill in Applies to, Used for, Name, Status, Priority and Confidence (a number from 0 to 1). Under Advanced settings, describe the rule in Advanced conditions and Advanced action. These use a structured format, and an example is shown to start you off. Advanced scope and Advanced source detail are optional. Select Create to save. If a structured field isn't in the right format, you'll see Check the JSON fields and try again.

Diagnostics

Diagnostics is an advanced tool for testing your rules before you rely on them. You give it a batch of example items in Advanced items, choose what the rules Applies to and what they're Used for, and select Run rules. The Resolver results show how many items were Resolved, Unresolved or Conflicted (where rules disagree).

Tick Use AI fallback to let the AI look at the items your rules couldn't resolve. Any new rules it proposes are listed under Proposed rules and saved in Approved rules with the status Proposed, so nothing takes effect until someone approves them. You can approve a proposed rule straight from the results.

Audit history

Audit history shows the 25 most recent AI suggestions and the decisions made on them, with each suggestion's name, Target type, status and Confidence. Use it to check what the AI has proposed and what your team approved or rejected.

Troubleshooting

• Check the JSON fields and try again. One of the structured fields isn't in the right format. Check brackets, commas and quotation marks, or start again from the example.
• No suggestions match the current filters. Clear the filters or select Refresh.
• No rules match the current filters. Clear the filters, or create your first rule with New.
• The AI stops with a limit message. Your package's monthly AI allowance has been used. See Package and add-ons.