If you have tried searching for a hyper-specific item on Amazon or Google Shopping recently, you know the frustration. You type in a nuanced request like “waterproof trail running shoes under $120 with a wide toe box,” and you are greeted with sponsored items, loosely related sneakers, or options that completely ignore half your criteria. San Francisco-based search startup Onton is aiming to fix this with the release of Ontology 1, a new model leveraging neurosymbolic AI search to deliver drastically more accurate e-commerce results.
In initial benchmark tests evaluated by independent LLM judges, Ontology 1 achieved a mean precision@10 score of 0.630, outperforming Google Shopping (0.543) and Amazon (0.469). Remarkably, it achieved these results while indexing only about 1% of the catalog size of those tech giants. But what makes this model so effective, and could it signal a major shift in how AI-powered product discovery works?
What Is Onton’s Ontology 1?
Ontology 1 is a purpose-built search model designed for complex, conversational, and multimodal product queries. Unlike pure large language models (LLMs) or traditional vector search engines, Ontology 1 combines two distinct artificial intelligence paradigms into what is known as a neurosymbolic system.
Neural networks excel at understanding natural language, nuance, and visual context from images. However, they frequently struggle with hard constraints, structured data, and precise logic—often leading to “hallucinations” or ignored query conditions. Symbolic AI, on the other hand, relies on explicit rules, knowledge graphs, and mathematical logic. By merging both approaches, Onton’s model uses deep learning to comprehend complex user intent while enforcing strict logic rules to verify that returned products actually meet every single search constraint.
Who Is It For?
While search technology ultimately impacts the end consumer, Ontology 1 is primarily positioned for e-commerce platforms, retail brands, online marketplaces, and enterprise search providers looking to upgrade their product discovery stack.
It is especially valuable for niche or complex retail verticals where technical specifications matter—such as consumer electronics, outdoor gear, specialized apparel, and home improvement. For shoppers, it promises an experience where conversational queries actually yield correct, highly relevant product suggestions on the first try.
Key Features of Neurosymbolic AI Search in Action
Onton’s Ontology 1 introduces several core capabilities that set it apart from standard vector-based search bars:
- Multimodal and Conversational Queries: Users can search using natural phrasing, upload reference images, or combine both in a continuous dialogue without losing track of constraints.
- Strict Logic and Constraint Satisfaction: The symbolic engine ensures that hard filters (such as price limits, dimensions, or specific materials) are never bypassed in favor of generic keyword matches.
- High Precision with a Smaller Index: By relying on deeper semantic understanding and structured reasoning rather than sheer brute-force indexing, the model delivers higher relevance even across smaller catalogs.
- Reduced Hallucination in E-Commerce: Traditional generative search models often recommend non-existent products or misread technical specs. Symbolic verification prevents these inaccuracies before results reach the user.
How Ontology 1 Compares to Google Shopping and Amazon
To understand why this model is generating buzz in the tech world, it helps to look at how traditional platforms handle search versus Onton’s neurosymbolic AI search approach.
Amazon’s search infrastructure remains heavily reliant on keyword matching, legacy indexing, and targeted advertising algorithms. As a result, exact searches often prioritize high-bidding sponsored products over items that strictly match the user’s prompt. Google Shopping utilizes advanced vector search and semantic AI, but can still struggle when queries contain multiple intersecting filters and specific logical conditions.
According to Onton’s 90-query benchmark scored by three independent LLM judges, Ontology 1 reached a precision score of 0.630 at top 10 results (precision@10). Google Shopping scored 0.543, while Amazon trailed at 0.469. This benchmark indicates that Onton’s hybrid architecture understands complex retail intent significantly better than standard industry search engines.
Pricing and Availability
At present, pricing is not publicly confirmed. Onton is currently deploying Ontology 1 through specialized enterprise partnerships and API integrations for e-commerce platforms. Organizations interested in testing or implementing the model generally need to reach out to Onton directly for custom quotes and developer access.
Our Verdict: Why Neurosymbolic AI Search Matters
At AI Tools Opinions, we have seen dozens of standard vector search tools enter the market promising “conversational shopping.” Most of them fall flat because pure LLM search tends to sacrifice precision for fluency. It sounds smart, but it gives you the wrong shirt size or ignores your budget limit.
Ontology 1 represents a necessary evolution in search architecture. By marrying the linguistic flexibility of neural networks with the non-negotiable precision of symbolic logic, Onton has tackled the biggest flaw in modern generative search. If Onton can scale this technology to massive global inventories, standard keyword search bars on major e-commerce sites will soon feel like relic technology. It is definitely an AI development worth watching closely for anyone in retail or web technology.
Frequently Asked Questions
What is the main advantage of neurosymbolic search over standard AI search?
Standard AI search relies heavily on statistical probability to guess what you mean, which can lead to missed details or inaccurate product specs. Neurosymbolic search pairs this broad understanding with strict, rule-based logic to ensure that logical conditions (like price, material, or exact size) are accurately enforced.
Can individual consumers use Ontology 1 directly on Onton’s website?
Ontology 1 is primarily an underlying search model built for integration into e-commerce stores, platforms, and applications. While demo environments may exist, it is designed as enterprise technology rather than a standalone consumer search engine like Google.
How does Ontology 1 handle image inputs?
The model uses multimodal visual processing to parse user-uploaded photos alongside natural language prompts. It can extract visual features (like color, pattern, or style) and combine them with written logic constraints to find matching products.