On August 28, the 2026 China International Big Data Industry Expo officially opened at the Guiyang International Conference and Exhibition Center. Hosted by the National Data Administration and organized by the People's Government of Guizhou Province, this year's Expo adopted the theme "Token — A New Path to Unlocking the Value of Data Elements."

As a leader in AI data infrastructure, KeenData showcased the KeenData Agentic Lakehouse Platform at Booth E1E06. The company officially launched its new Keen Agentic OS intelligent agent development operating system alongside the upgraded KeenData Lakehouse multimodal lakehouse platform, offering a comprehensive view of its product capabilities and engineering practices spanning AI data foundations, model training and inference, and agent operations.

The AI industry is shifting its focus from large-scale model training toward the scalable deployment of intelligent agents. However, agent development across industries continues to face bottlenecks — data integration challenges, complex compute scheduling, and high customization costs — making widespread deployment difficult. Drawing on hundreds of agent project implementations and real-world business scenarios, KeenData officially launched Keen Agentic OS, an enterprise-grade agent-native development operating system.

The launch program spanned the full day. In the morning session, KeenData provided a systematic overview of the product's positioning, technical architecture, and deployment roadmap, complemented by a live demonstration of multi-agent collaboration executing complex business tasks — showcasing the complete chain from task decomposition and autonomous decision-making to coordinated execution. In the afternoon technical deep-dive, experts from KeenData walked through each core capability module, engaging in in-depth discussions with attendees on pain points and solutions across various business scenarios. Together, the two sessions conveyed KeenData's product proposition of "driving agent deployment at scale through AI-powered data," sparking active on-site engagement.

At the technical architecture level, Keen Agentic OS is built around six core dimensions:

Data+Agent Deep Integration connects enterprise data and business workflows, bringing this year's "Token" concept into the agent runtime to unlock the value of enterprise data assets. The Ontological Semantic Knowledge Base transforms scattered enterprise data into intelligence assets that agents can understand, reason with, and invoke. State-of-the-Art Training and Inference Performance enables unified multi-model management, dynamic scheduling, and elastic inference. An end-to-end closed-loop approach completes the "Data-Action-Operations" chain. The Powerful Agent Runtime delivers integrated capabilities for context understanding, task planning, workflow execution, memory management, and tool orchestration, enabling agents to execute production tasks with stability, continuity, and autonomy. Multi-Agent Collaboration allows multiple specialized agents to work together autonomously, forming AI teams to tackle complex business challenges. Enterprise-Grade Governance and Securitycovers permissions, compliance, observability, evaluation, and lifecycle management, ensuring agents remain controllable, auditable, and manageable. Together, these capabilities form a closed loop that significantly reduces agent customization costs and compute management complexity, enabling enterprises to rapidly deploy and scale production-grade agents.

Alongside Keen Agentic OS, KeenData also unveiled a major upgrade to the KeenData Lakehouse multimodal lakehouse platform. Powered by the AI-in-Lakehouse native intelligent architecture engine, the platform addresses traditional data foundation gaps in agent development and token management through advancements in multimodal data processing and training-inference acceleration.

On the multimodal data processing front, the AI-in-Lakehouse engine builds an integrated pipeline within the lakehouse, handling image, video, audio, and text parsing, cross-modal alignment, token extraction, and full-chain lineage governance in a single workflow. This directly produces trusted token assets and eliminates knowledge silos across multiple agents. On the training-inference acceleration front, the AI-in-Lakehouse engine orchestrates the KMI training-inference engine, establishing a high-speed GPU direct-connect link that reduces data transfer overhead and boosts GPU throughput, continuously supporting incremental token iteration and enabling high-concurrency, low-latency multi-agent collaboration. The AI-in-Lakehouse architecture provides the intelligent foundation; multimodal capabilities address token quality; the training-inference engine addresses deployment speed; and the Harness engineering system addresses large-scale delivery — together ensuring the data foundation runs efficiently as a reliable supply engine for agent operations at scale.

By incorporating "Token" into its core theme, this year's Expo signals a fundamental shift in how data element value is measured — moving from "how much data do you have" to "how much data is being consumed by AI." Token is becoming the new metric for data value creation. Keen Agentic OS focuses on the core challenge of scaling agent deployment across the full lifecycle of development, operations, and governance. KeenData Lakehouse addresses the underlying bottleneck of ensuring a stable supply of high-quality tokens. Together, these two platforms form the KeenData Agentic Lakehouse Platform, delivering a unified AI data foundation that integrates data infrastructure, multimodal computing, agent development, and training-inference capabilities — providing enterprises with an integrated solution for production-grade agent deployment and unlocking real business value from AI.

The capabilities of the AI data foundation ultimately prove themselves in real-world industry scenarios. During the Expo, Gao Hailing, Vice President of KeenData, participated in a closed-door session on "Public Resource Trading Data Quality Improvement and Analytics Applications," engaging with industry leaders on data standardization, multi-source data integration, and full lifecycle governance.

KeenData's booth also featured deployment case studies across government, energy, manufacturing, and other core industries, with capabilities already validated across more than 300 large enterprises. On the ecosystem front, KeenData has completed deep compatibility and integration with leading domestic chips and large language models, building a comprehensive "National Chip — National Model — National Data" ecosystem to advance autonomous AI infrastructure in China. The booth also featured a Token-based elastic pricing experience zone, allowing visitors to assess the return on investment for scaled agent deployment and providing a quantifiable reference for future collaboration.

The 2026 Big Data Expo runs through August 30. KeenData invites all partners and industry professionals to visit Booth E1E06 at the Guiyang International Conference and Exhibition Center.