Pinecone Launches Nexus Engine: Integrating Business Context for AI Agents and Generating Structured Data

Pinecone believes that large language models excel at general knowledge, and vector databases can locate specific information in scattered files, but enterprises truly rely on business context information—dispersed across contracts, knowledge bases, HR documents, meeting notes, customer service tickets, and financial records. Although agents can search this information, retrieving it for every task is highly inefficient, leading to high token costs, slow responses, and potentially missing information in outputs.

Nexus integrates scattered enterprise knowledge into a structured layer that agents can query directly. Token consumption shifts from per-query retrieval cycles to a one-time data structuring phase. Early adopters have seen significant performance improvements, especially in financial services and legal research.

In the legal domain, Nexus completes all assigned tasks, while coding agents achieve only 6% and RAG systems 66%. RAG systems perform poorly on 'doctrine synthesis, cross-case reasoning, and coverage issues'—tasks requiring integration of information from multiple sources. Pinecone data shows token consumption is approximately 1/9 to 1/15 of original.

In enterprise data management, Nexus achieves 90% accuracy vs. 65% for RAG, with a cost of $0.0038 per document processed.

Nexus uses Workspaces as top-level containers, typically corresponding to teams or business units. Data is divided into Contexts, each corresponding to a specific dataset or knowledge domain. Manifests define how raw data sources are ingested and transformed into structured knowledge.

Domain experts can pre-design blueprints defining entities and their relationships, solidifying domain knowledge into the data structuring layer before queries. Agents do not need to parse the corpus structure during query time; they directly follow the expert's structuring logic.

Data ingestion is achieved through connectors, supporting local files, Box, and Microsoft OneLake. Pinecone states that support for Google Drive, Slack, GitHub, Notion, Confluence, and S3 is coming soon. After data is structured, it can be queried via KnowQL, usable by agents, chatbots, and recommendation systems.

Nexus offers a preview sandbox where users can connect data sources, design contexts, and run queries to validate feasibility. It also provides a BYOC (Bring Your Own Cloud) deployment option for environments with mandatory data residency, security, and compliance requirements.

Existing similar products include Cognite, RationalAI, LlamaIndex, among others. Nexus's early data comes from Pinecone's own reports and requires independent verification. Specific performance improvement and cost reduction figures (e.g., completion rates, accuracy, token consumption ratios) are claimed by Pinecone and have not been third-party confirmed. Additionally, Nexus's generalization ability for complex cross-domain tasks and stability at scale have not been publicly disclosed.