Pinecone Newsroom

Pinecone Assistant now has support for new OpenAI (gpt-4.1, o4-mini) and Anthropic (claude-3-7-sonnet) models, as well as Gemini (gemini-2.5-pro). In addition to the new selection of models, Pinecone Assistant now gives developers more control by exposing the temperature parameter for the chosen LLM.

Pinecone sponsored and attended events across Tech Week by A16Z

This blog post explores how multi-vector retrieval improves search accuracy by capturing rich query-document interactions, while addressing its scalability challenges. It introduces a practical, staged retrieval pipeline that balances speed and effectiveness, starting with fast retrieval, refining with multi-vector embeddings, and finishing with cross-encoder reranking. The post highlights ConstBERT, a constant-space multi-vector model co-developed by Pinecone and academic collaborators, and shows how to integrate it into Pinecone to build efficient, scalable, and accurate search systems. ConstBERT is now available in open source.

We recently partnered with the School of Code during their AI Week. The students used Pinecone to build working AI applications in just a few days. A short video captures their projects and impressions. While the apps are indeed beginner-level, they show that even newcomers can build practical AI experiences with modern tools. The signal is clear: Pinecone is not just for enterprises; - it’s for anyone who wants to build knowledgeable AI -- today.

As AI systems become agentic and more autonomous, enterprises need infrastructure that delivers secure data access, real-time retrieval, and production-grade performance. Rubrik Annapurna, powered by Pinecone’s vector database, provides the infrastructure to deploy generative AI at scale, with RAG and agentic workloads in mind.

We are releasing our first three MCP servers: Pinecone Assistant MCP (remote), Pinecone Assistant MCP (local), and Pinecone Developer MCP (local). Together, the three support the following use cases: Providing structured context for agents (remote and local Pinecone Assistant MCP servers as well as local Pinecone Developer MCP server), enabling in-IDE documentation search (local Pinecone Developer MCP server), and allowing developers to perform vector database operations directly within popular development environments (local Pinecone Developer MCP server).

Pinecone has launched the first remote MCP server to bring knowledge to agentic workflows. Starting today, every Pinecone Assistant is now also an MCP server! Users can now connect their Assistant as a context tool in their agents and AI applications. This means that developers can now combine the simplicity of Assistant document processing, contextualization, and agentic-RAG with the simple integration and ecosystem of MCP.

We're rolling out new features each day this week (March 17-21). Stay tuned for updates!

This post describes what we mean by agentic workloads, their patterns, and how our new system optimizes for them. We also describe how we've improved performance for our search and recommendation systems workloads.

Explore how our optimized embedding and reranking inference uniquely enhances retrieval quality, simplifies the developer experience, and reduces the operational footprint of modern retrieval applications.