Octen, a startup with software that enables artificial intelligence agents to search the web, launched today with $10 million in seed funding. 

Square Peg led the investment. It was joined by Singapore-based venture capital firm Argor and a group of AI researchers.

An AI agent that requires the ability to browse the web can theoretically do so by entering queries into one of the many free search engines out there. However, that approach comes with performance limitations. The reason is that off-the-shelf search engines are usually geared toward humans, who can’t review data as quickly as an AI model.

Octen, officially APITECH AI Pte Ltd., is working to address that challenge. The three-month-old startup has built a search engine specifically for AI agents. It says the service, which is accessible through an application processing interface, is the fastest of its kind on the market by a wide margin.

The company measured its search engine’s speed using a benchmark called SealQA Hard. It comprises more than 200 multistep questions that require AI models to retrieve data from the web and use reasoning methods to analyze it. Octen’s search API answered the questions with a median response time of 62 milliseconds, or more than four times faster than its speediest rival.

AI agents that use Octen can further speed up searches by breaking them up into multiple queries. The service can run those queries side-by-side, which is significantly faster than completing them one after another. Octen says its infrastructure can process more than one million queries per second per customer account.

Google LLC has implemented a similar parallelization method in its search engine’s AI Mode. When consumers ask a complex question, the feature uses a technique called query fan-out to run multiple searches simultaneously. It then distills the data retrieved by those queries into a single natural language response.

Octen says its API lends itself to not only web searches but also other other data retrieval tasks. A financial institution, for example, could use the service to generate stock performance reports based on publicly available data.

AI agents turn webpages and other files into mathematical representations called embeddings before analyzing them. Octen offers its flagship search engine alongside a series of cloud-hosted algorithms optimized to generate embeddings. The most capable of those algorithms, Octen-Embedding-8B, has set a record on an AI data retrieval benchmark called RTEB.

“An agent can process thousands of data points simultaneously, which places a massive technical burden on the underlying stack,” said founder and Chief Executive Kuan Zou. “By solving the challenges of sub-100-millisecond latency and million-level QPS, Octen is enabling a new milestone in AI applications by allowing agents to reason over the live web with the same speed and fluidity as memory.”

Octen has used its seed funding to build out its engineering and developer relations teams. The company, which maintains offices in Singapore and San Francisco, has also bought infrastructure for the globally distributed backend that hosts its search engine.

The search API is currently available through an invite-only beta program. The company says the service is being tested by multiple AI software providers.

Image: Unsplash