Pinecone launches Nexus knowledge engine for AI agents

Pinecone has made its Nexus knowledge engine generally available. The product is designed to give enterprise AI agents governed access to knowledge drawn from proprietary data.

The launch comes as businesses try to move AI agents from pilots into regular operations while facing rising computing costs and questions about how sensitive internal knowledge is handled. Pinecone argues that many existing retrieval systems were built for human users who could manually check outputs, not for software agents operating autonomously.

Knowledge layer

Nexus is positioned as a knowledge layer between a company's internal data and the AI models its agents use. It compiles data and workflows into domain-specific knowledge that agents can query through a language called KnowQL, rather than reconstructing context from raw data with each request.

The system can be deployed in a customer's own cloud environment and used with different model providers, including open-weight models. It also includes field-level access controls, per-field citations, confidence scores, lineage back to source material, and controls for personally identifiable information during ingestion.

Pinecone says this approach addresses two problems that have become more visible as companies expand their use of AI agents. One is the cost of repeatedly retrieving and reprocessing information before an agent completes a task. The other is concern that feeding proprietary data into external model systems could weaken a company's competitive position if those same models are widely available to rivals.

Benchmark results

The company tied the launch to benchmark results on τ-Knowledge, an open benchmark created by Sierra for agentic customer support work. According to Pinecone, an agent using Nexus as its knowledge layer scored 47.4%, ahead of 46.4% for GPT-5.5, which it described as the best frontier model currently listed on the leaderboard.

Pinecone also said the same setup reduced cost per task by 74% compared with an agent using a frontier model without the Nexus knowledge layer. The benchmark is intended to test multi-step reasoning, policy adherence, and coordinated tool use, and measures whether an agent reaches the correct end state in customer support tasks.

The company made broader claims about the product's economics and performance beyond the benchmark results. It said precompiled knowledge can cut token costs by more than 90% compared with agentic retrieval-augmented generation, produce answers up to 30 times faster, and complete tasks with more than 90% accuracy.

Those claims go to the heart of a broader debate in enterprise AI over whether model performance alone is enough to improve business processes. Model providers continue to compete on scale and reasoning, but software companies building enterprise systems increasingly argue that the quality, structure, and governance of proprietary knowledge will play a larger role in real-world deployments.

Enterprise focus

Pinecone said Nexus depends on subject matter experts to shape the knowledge used by agents rather than relying on static central ontologies. It said this is intended to align the system with business context and day-to-day workflows in areas such as finance, insurance, legal services, sales, and customer support.

That emphasis also reflects a shift in how AI software is sold inside organisations. Earlier enterprise AI tools were typically built for developers and technical teams, but suppliers are now trying to appeal to line-of-business users expected to guide how AI is embedded in specific tasks and processes.

"Enterprises adopting AI are squeezed from two sides," said Ash Ashutosh, Chief Executive Officer, Pinecone. "Agents burn tokens grinding through raw data, so cost and latency climb while accuracy stays lower than it should be. And every model call risks handing proprietary knowledge to a system that can turn around and compete with you. Nexus puts a knowledge engine in your own cloud, raises accuracy, lowers the total cost of running AI, and keeps your own experts shaping how agents work."

Industry outlook

Industry analysts have also been examining whether retrieval costs are emerging as one of the main constraints on agent deployments. Companies running large numbers of autonomous tasks can spend heavily on repeatedly fetching, filtering, and reformatting information before a model produces an answer or takes an action.

"Enterprises running agentic workloads have been hitting a real ceiling on cost, since retrieval and re-orientation can eat up the bulk of token spend before an agent ever reasons," said Devin Pratt, Research Director, IDC. "Pinecone's approach, compiling proprietary knowledge into a reusable layer instead of re-deriving it on every call, is a sensible response to that problem. It's a promising direction, and one worth watching as more enterprises evaluate precompiled knowledge layers."

Nexus sits within Pinecone's broader platform alongside its database product and a marketplace for knowledge applications. Pinecone said it serves more than 10,000 customers and 1 million developers worldwide.