Nexla Open Sources its Agentic Chunking Technology to Improve AI Accuracy for All

Agentic chunking contribution is first step in open sourcing agentic AI technology to spur innovation and accelerate enterprise-grade GenAI

SAN MATEO, Calif. – March 18, 2025 (GLOBE NEWSWIRE) — Nexla, a leader in AI-powered integration for data and AI, is contributing key innovations from its cutting-edge agentic AI framework to the open source community, reinforcing its commitment to advancing enterprise-grade AI technology. Building on years of technological leadership, Nexla has released its groundbreaking advancements in agentic chunking to developers worldwide, empowering organizations to create more accurate GenAI-powered agents and assistants while accelerating industry-wide innovation.

Nexla has the industry’s first AI-powered integration platform that handles today’s overwhelming data variety, replacing endless connectors, diverse formats, and infinite schemas with AI-ready data products. With Nexla, you can integrate any document, data, app, or API, create AI-ready data products, and deliver GenAI projects without coding, up to 10x faster than the alternatives.

Nexla uses AI to connect, extract metadata, and transform source data into human-readable data products, called Nexsets, that can be shared in a built-in marketplace for true data reuse and governance. Nexla’s agentic AI framework lets companies implement agentic RAG for agents and assistants without coding, using LLMs during each stage to improve accuracy. For example, Nexla can get context from multiple data products, use a unique algorithm to rank, prioritize, and eliminate data, and then combine the context with a rewritten query and submit it to just about any LLM.

Agentic chunking represents the next evolution of document processing for Retrieval-Augmented Generation (RAG), providing AI engineers with an intelligent, structured, and scalable way to break down complex documents for optimal retrieval and generation.

Agentic Retrieval –– Retrieval refers to the moment during RAG inferencing when context is retrieved. Unlike most other products, which only retrieve data from a specific source, like a vector database, Nexla uses an LLM to dynamically find data across any relevant data products while enforcing data security policies. Expanding the sources of data using Nexla’s agentic retrieval can dramatically improve AI accuracy without having to load all the data into a vector database.

Nexla Orchestrated Versatile Agents (NOVA) –– NOVA, which has been in use by early adopters, is now generally available. NOVA is an agentic interface within the Nexla UI, built on Nexla’s agentic AI framework that lets anyone compose end-to-end pipelines by using natural language (e.g. English) commands. Nexla uses an LLM to create the tasks and code (Python and SQL) to query sources, build transforms, and orchestrate pipelines. Users can work entirely within NOVA or move between NOVA, no-code point-and-click configuration, and pro-code mode.

Agentic chunking has delivered the following benefits over other chunking techniques across internal tests and production deployments:

• Smarter document understanding: Instead of blindly splitting text into fixed-sized chunks, agentic chunking treats documents as structured knowledge, identifying key sections, headings, and relationships.
• Precision-driven efficiency: Uses LLMs like GPT-4o only where they add real value—detecting and classifying headings—while relying on smaller models and deterministic rule-based processing for everything else to achieve the best price-performance.
• Improved retrieval and accuracy: By preserving hierarchical relationships and semantic structure, chunks retain essential context, leading to significantly better responses from RAG-based systems.
• Enterprise-grade: Scales linearly with document size and has proven its reliability in production deployments.
• Domain adaptability: Incorporates domain-specific chunking strategies beyond generic embeddings, ensuring AI-powered retrieval works optimally for financial reports, technical manuals, legal documents, and more.

“Companies who have deployed GenAI assistants and agents often cite data quality and AI accuracy as two of their top challenges. They’re related: bad data leads to bad outcomes and AI hallucinations,” said Saket Saurabh, Nexla Co-founder and CEO. “Our agentic AI framework has dramatically improved AI accuracy and scale for our customers. But we believe the best solution is to solve these problems together as an industry by jointly contributing to open source. Open sourcing agentic chunking is just the first step. We’re excited to work with other vendors, and to release more of our agentic AI technology to help companies get to even higher quality data and outcomes.”

To learn more about Nexla, visit us at GTC 2025 in San Jose at booth No. 2008, or visit:

• Nexla Integration Platform
• Nexla Customer Success Stories
• Data Integration Resources
• Nexla Open Source Agentic Chunking (GitHub)

About Nexla 

Nexla is a leader in AI-powered integration for data and AI whose mission is to make data ready-to-use for everyone. The Nexla Integration Platform is the first integration platform powered by AI and built to handle today’s data variety. With Nexla you can integrate any data, create AI-ready data products, and deliver enterprise-grade GenAI without coding, up to 10x faster than the alternatives.

Trusted to deliver mission-critical data by leading companies including DoorDash, LinkedIn, Johnson & Johnson, and LiveRamp, Nexla is headquartered in San Mateo, California. It has been recognized in the 2022, 2023, and 2024 Gartner Magic Quadrant™ for Data Integration Tools and top-rated by customers on Gartner Peer Insights.

Media Contact
BOCA Communications
nexla@bocacommunications.com