Pinecone Launches Nexus Engine: Structuring Business Context for AI Agents
Pinecone has introduced Nexus, a "knowledge engine" for AI agents that transforms enterprise data into a structured layer. It reduces token costs and improves accuracy, achieving 90% accuracy in tests compared to 65% for RAG, with token consumption ranging from 1/9 to 1/15 of the usual amount.
Pinecone has announced the public release of Pinecone Nexus, a knowledge engine designed for AI agents. Nexus transforms disparate corporate data (contracts, knowledge bases, documents, meetings, tickets, finances) into a structured layer that agents can query directly. Unlike RAG, expensive search is not performed for each query — token costs are shifted to the one-time data processing stage. Early adopters in finance and legal sectors reported significant improvements: Nexus completed 100% of legal tasks, compared to 6% for a code agent and 66% for RAG. Token consumption decreased by 9–15 times. In enterprise data management, Nexus achieved 90% accuracy versus 65% for RAG, with a processing cost of $0.0038 per document. The key concept is a Workspace, within which data is divided into Contexts. A Manifest defines rules for transforming raw sources into structured knowledge; it can incorporate domain-specific expertise. Data is connected via connectors (local files, Box, Microsoft OneLake; soon Google Drive, Slack, GitHub, Notion, Confluence, S3). Queries to structured data are made using KnowQL. A preview mode is available to check data schemas, and a Bring Your Own Cloud (BYOC) option supports storage and security requirements. Competitors include Cognite, RationalAI, and LlamaIndex.
Source: InfoQ 中国 —
original