AI Assistant
The AI Assistant is a conversation with an AI model about your rules, inside the Rule Repository. You ask in plain language and it does the work with the same tools you use: it finds rules and sources, explains what a rule does and why, writes new rules, drafts and tests, runs tests and looks at how rules behaved in production. It can answer questions such as "Which rules use this field?", "Why did this order get held?" or "Write a test for this rule".
The Assistant always acts for you and within your permissions, under stricter limits than a person has: it never makes anything live, never approves or declines, and never deletes. Everything it saves is saved as a request for approval, so a person always reviews its work before it reaches production.
The AI Assistant is available only when your application has connected at least one AI model and you have permission to use it. Your application decides which models are offered.
The Assistant uses these parts of the repository:
- The AI button, at the top right of the screen beside + New. It opens the AI Assistant dialog.
- The header of the dialog: the conversation's title and a list of models (the model that answers your next message). If you have permission to read other users' conversations, a My conversations list sits before it.
- The left column: + New Conversation, then your conversations, newest first, grouped by day. Hover over a conversation to Rename or Delete it. If you have permission to see usage, two tabs sit over the list: Conversations and Usage.
- The conversation in the middle: your messages, the AI's replies, and cards for things that need you or that it saved.
- The message box at the bottom, with a paperclip to Attach a file (when your application allows attachments) and Send. While the AI works, Stop takes the place of Send.
- The bar under the message box: the conversation's token count on the left; Detailed answers and Steps on the right.
- The Steps pane, which slides in from the right and lists, in plain words, each thing the AI did in a turn: "Looked up rules", "Read a rule", "Ran a test".
Starting a Conversation
- Click AI at the top right of the screen. The AI Assistant dialog opens with the conversation you had open last, or an empty one.
- To start fresh, click + New Conversation.
- Optionally, pick the model in the list at the top right.
- Type your question or request in the message box, for example "What does the rule that prices orders do?". Press Enter or click Send. (Shift+Enter starts a new line.)
- While the AI works, the conversation shows Working.... When it finishes, its reply appears, with a line under it: how many steps it took, the tokens it used and the model that answered.
The conversation's title is taken from your first message. To change it, hover over the conversation in the left column and click Rename.
Following What the AI Does
- Check Steps under the message box to keep the Steps pane in view. Each step is listed as it happens.
- Or click the "steps" link under any reply to see that turn's steps for a few seconds.
- Uncheck Detailed answers to get only the essential answer, without explanations. It applies from your next message, and the repository remembers your choice.
Answering the AI's Questions
Some actions need your answer before the AI can go on. They appear as cards in the conversation:
- Before a change to an existing rule, a card shows what the rule says today and what it would say after the change. Nothing is saved until you say yes.
- Before a test that would call your application's real methods, a card lists those calls and asks whether to run them.
- When the AI reaches its step limit for one turn, a card offers Continue.
Click the card's button to answer. The AI cannot answer these questions for you.
Opening What the AI Saved
When the AI saves a draft, a rule version or a test, a card in the conversation says so and offers Open. Click it to open what was saved in its own dialog, where you can review it, test it and, if you have permission, approve it.
Attaching Files
If your application allows attachments, you can give the AI a document to work from, such as a policy that the rules should follow.
- Click the paperclip in the message box and choose the file.
- Type your request and click Send.
Not every model reads every kind of file. If the selected model cannot read yours, a card says so and names the models that can; pick one of them at the top and send again.
Stopping the AI
Click Stop while the AI is working. The step under way finishes, and the turn ends with "Stopped." Send a new message to go on.
Going Back to Earlier Conversations
Your conversations are listed in the left column. Click one to open it and continue where you left off. To delete one, hover over it, click Delete and confirm; a deleted conversation cannot be brought back. Conversations nobody has touched for a while are removed automatically; your application decides how long they are kept (90 days unless it says otherwise).
Reading Other Users' Conversations
With permission to read every user's conversations:
- Open the My conversations list at the top and select a user. Their conversations appear in the left column.
- Click a conversation to read it. A band says whose conversation it is: nothing can be changed, and reading it is not recorded anywhere.
- To carry on from it yourself, click Continue as me. The conversation is copied into your own, under a title that starts with "Continued from", and the original is left as it was.
- Select My conversations again to return to your own.
Seeing Usage
With the same permission, the Usage tab beside Conversations shows how much the AI has been used: for a period you choose (Today, Last 7 days, Last 30 days or Last 90 days), a table with each user and model, the number of turns, the tokens used, the part of them read from the model's cache, and the cost when your application has given the model's prices.