Pinecone Launches Nexus to Give AI Agents Enterprise Knowledge

Modern AI agents often struggle with the inefficiency of re-deriving context from raw data during every interaction. According to Pinecone, over 85% of large language model effort is currently wasted on repetitive retrieval, which drives up latency and consumes excessive tokens. Nexus addresses this by moving knowledge management into a purpose-built layer that deploys directly within a company’s own cloud environment, ensuring that sensitive data remains under internal governance rather than being exposed to external model vendors.

In testing on Sierra’s τ-Knowledge benchmark—a standard for multi-step reasoning and policy adherence—agents integrated with Nexus achieved the highest scores, surpassing standalone models from OpenAI, Anthropic, and Google. Beyond performance gains, the system offers a substantial financial advantage. Data indicates that adding the Nexus layer can reduce task costs by up to 81% for models like GPT-5.5. By utilizing KnowQL, a declarative query language, the system allows subject matter experts to shape how agents access business context, effectively turning Pinecone from a developer-focused database into a broader platform for line-of-business professionals.