Chat Search

Type a shopping question on the homepage chat (for example, "a thoughtful birthday gift for a coffee lover under $60"). Biasque runs that query through an AI agent that searches public product catalogs, evaluates candidates, and returns a short list of recommendations along with a written reason for each pick.

Every search, every product the agent considered, and every reasoning trace is stored. That data is what powers the dashboards, the trend charts, and the Monitor API.

The agent behaves like a careful shopper, not a keyword matcher. Treat it that way:

• State the use case, not just the product type. "Lightweight rain jacket for cycling commutes in Portland" beats "rain jacket".
• Include constraints. Budget, size, color, materials to avoid — anything you would tell a salesperson.
• Mention the recipient."For my dad who already has everything" produces very different picks than a bare product name.
• One ask per query. If you have two unrelated needs, run them as two separate searches so the reasoning stays focused.

For each recommended product you will see:

• Product card.Title, image, store, price, and a link to the live product page on the merchant's site.
• Agent reason.A short explanation, in the agent's own words, of why this product fits your query. This is the data point that does not exist anywhere else.
• Selected vs. seen.Some products are surfaced in the search but not chosen. Both states are logged so trend analysis can tell the difference between "invisible to AI" and "visible but never picked".

When you run a search, we store the candidate products the agent considered, which ones it selected, and the agent's reasoning for each pick. Your raw chat input is kept private and visible only to you. The AI-generated search query the agent uses internally may appear anonymously in aggregate trend data (with no link to your account). Signed-in merchants can also access richer breakdowns through the dashboard and the Monitor API.

For details on data handling, see the Privacy Policy.

• Single model perspective. All searches are processed by one underlying AI model. Results reflect how that specific model evaluates products — not a consensus across multiple AI systems. Selection rates and reasoning data are best read as signals from one AI perspective, not a universal truth.
• Catalog coverage. The agent searches public catalogs exposed through the Shopify Catalog MCP and similar feeds. Stores not in those feeds will not appear.
• Freshness. Prices and stock shown in search results reflect the most recent catalog snapshot. The live product page is the source of truth.
• Rate limits. All searches are subject to rate limits to keep the service responsive. Anonymous users are limited per IP; signed-in users have a daily allowance. If you hit a limit, wait a few minutes or try again the following day.