AI context & examples
Requires the Grow plan or higher.
Detection tells the bot what your fields are. The AI Context and AI Examples tabs tell it how to use them — your business rules and worked examples. Find both on the catalog detail page (Dashboard → Product Catalogs → Manage).
The use-case hint
The simplest steering signal is the When should AI use this catalog? description you set when creating the catalog (editable on the Config tab). When a tenant has more than one catalog, the bot uses this hint to choose the right one — for example, routing "show me apartments" to a real-estate catalog rather than a parts catalog.
AI Context tab
The AI Context tab holds Business Rules — free-form, domain-specific instructions (up to 4,000 characters) injected directly into the product search tool's description. Use it for conventions and shortcuts the bot can't infer from raw data, such as:
- Data conventions — value formats and casing (e.g. city names are uppercase; is an integer wheretext
propertyType).text1 = apartment - Lexicon — map customer words to filter values (e.g. "kuca" → ).text
propertyType=2 - Business rules — query shortcuts (e.g. "cheapest" → sort by price ascending; "luxury" → minimum price 200,000).
Pick a starter from Load template (Real Estate Serbia, E-commerce, Restaurants) or write your own, then click Save Business Rules. Saving regenerates the search tool so the bot picks up the change immediately.
The Available filters panel above shows the enum labels the AI sees for each filterable field. These are generated automatically during sync; click Rebuild AI Context to regenerate them, or expand Show advanced to edit labels and data observations by hand.
🖼️ [Image] — The AI Context tab with Business Rules editor and Available filters panel
AI Examples tab
The AI Examples tab holds few-shot examples that guide the AI pipeline at four stages:
- Classifier — intent-classification examples.
- Planner — tool-plan generation examples.
- Synthesis — response-writing examples.
- Verifier — anti-fabrication checks.
Switch stages with the tabs and filter by language, difficulty, or source. Enabled examples are injected into that stage's prompt, nudging the bot toward the patterns you want for tricky or recurring queries. Promptly auto-generates a starter set; you can edit, enable, or disable each one.
🎬 [Video] — Adding business rules and reviewing planner few-shot examples
Together, the use-case hint, business rules, and few-shots shape how the bot reasons over your catalog — see Multi-step reasoning and how the search tool itself is built in Custom tools & API actions.