A small index, a big precision lead
Onton, a San Francisco-based search and discovery company, has released Ontology 1, a neurosymbolic model for complex, conversational, multimodal product search. On a 90-query benchmark scored by three independent LLM judges, Ontology 1 reached a mean precision@10 of 0.630, against 0.543 for Google Shopping and 0.469 for Amazon. It hit those numbers while indexing roughly 1% of their catalogs.
The result is meaningful because the queries that Onton targets are the kind of messy, requirements-heavy prompts that current e-commerce search handles badly. A user who types "pet-friendly sectional that fits a 9-foot wall and ships in two weeks" is not asking for a category match. They are asking for a constraint intersection that traditional faceted search cannot represent.
Conventional e-commerce search, Onton argues, has barely changed in nearly 30 years. It assumes intent maps onto attributes: size, price, material, brand. There is no filter for pet-friendly, and no filter for furniture that fits the room.
How the model works
Ontology 1 takes a different approach. For "pet-friendly sectional," it does not trust the seller's label, which may be a marketing term with no real definition. It checks the listing body for actual pet-relevant signals: removable cushions, stain-resistant fabric, machine-washable covers, customer reviews mentioning pets. The result is a candidate set weighted by grounded signals rather than seller-declared ones.
The system is also designed to handle negation, which is the failure mode most current search engines get wrong. A query for "leather sofa, not brown" should exclude brown sofas. Most search engines treat the "not brown" as either ignored or applied too aggressively. Ontology 1 has been trained to handle these cases explicitly, which is what moves the precision needle on the benchmark.
Who gets to use it
Ontology 1 is live for end users at onton.com, and the company is granting partner access case by case for teams building on what it calls the agentic web. There is no public API, no pricing tier, and no open checkpoint for the model itself. Adoption today looks like a partnership, not a product.
The addressable market is narrow but real: mid-market and enterprise retailers, marketplaces, and agentic commerce platforms whose relevance stack already loses on long, requirements-heavy queries. Small catalogs see less benefit, because the failure mode Ontology 1 targets scales with catalog size and listing noise. The current vertical focus is home decor and furniture, though Onton says the methodology generalizes beyond e-commerce and that the model can search non-product data with essentially no reconfiguration.
The interesting question is whether retailers will treat this as a feature to license or as proof that the search problem is finally tractable enough for someone to disrupt the incumbent tooling.