HyT Capital Portfolio | KeenData Launches New Product Keen Agentic OS
HyT Capital's early-stage portfolio company, KeenData — a leader in AI data infrastructure — has officially launched its new product, the Keen Agentic OS agent development operating system, along with the upgraded KeenData Lakehouse multimodal lakehouse platform. The release systematically demonstrates the company's product capabilities and engineering practices across key stages, from AI data foundation and model training to agent operations.
--
Keen Agentic OS Officially Launches,
Exploring New Pathways for the Value of Data as a Production Factor
Currently, the center of gravity in the AI industry is rapidly shifting from large model training to large-scale Agent deployment. However, the development of industry-specific Agents is widely constrained by challenges such as difficult data adaptation, complex compute scheduling, and high scenario customization costs, making scaled deployment difficult to achieve. Drawing on hands-on experience from hundreds of Agent projects and refinement through massive real-world business scenarios, KeenData has officially launched Keen Agentic OS, an enterprise-grade Agent-native development operating system.
During the new product release, KeenData systematically articulated the product's positioning, technical architecture, and implementation roadmap, and presented a full-process demonstration of multi-agent collaboration in completing complex business tasks — visually showcasing the complete chain from task decomposition, autonomous decision-making, to coordinated execution — conveying KeenData's product proposition of "driving large-scale Agent deployment with AI data." Keen Agentic OS:
An Enterprise-Grade Agent Development Operating System
At the technical architecture level, Keen Agentic OS systematically builds capabilities across six core dimensions:
At its core, it integrates deep integration of Data and Agent, connecting enterprise data and business workflows to bring the concept of "tokenization" to life during Agent runtime and revitalize enterprise data assets. It features an ontology-based semantic knowledge base that leverages enterprise ontology as the core, integrating multimodal knowledge construction, business semantic modeling, hybrid retrieval, and context generation capabilities — transforming scattered enterprise data into intelligent cognitive assets that Agents can understand, reason over, and invoke. It supports extreme training and inference performance through unified multi-model management, dynamic scheduling, and elastic inference. It targets an end-to-end closed loop, connecting the full chain from data to action to operations. The powerful Agent Runtime provides an Agent-native runtime environment that integrates context understanding, task planning, workflow execution, memory management, and tool orchestration — enabling Agents to operate with stability, continuity, and autonomy. Multi-Agent collaboration serves as the guiding direction, supporting multiple specialized Agents to autonomously divide labor and coordinate on complex business tasks, forming AI teams to tackle difficult problems. Enterprise-grade governance and security ensure coverage of access control, security compliance, runtime observability, effectiveness evaluation, and lifecycle management — keeping the Agent system controllable, manageable, and auditable. These capabilities form a complete closed loop, significantly reducing the cost of Agent customization and the complexity of compute management, truly supporting the rapid deployment and scalable, controllable operation of production-grade Agents in the enterprise.
KeenData Lakehouse Iteration:
Filling the Data Foundation Gap in Token Supply
Alongside the launch of Keen Agentic OS, KeenData also released a major iteration of its multimodal Lakehouse platform. The platform's core uses an AI-in-Lakehouse native intelligent-driven architecture engine, addressing two key directions — multimodal data processing and training-inference acceleration — to close the capability gaps in traditional data foundations for Agent development and token management.
On the multimodal data processing front, the AI-in-Lakehouse engine builds an integrated processing pipeline directly within the lakehouse, completing in one go the parsing of images, text, audio, and video, cross-modal alignment, token extraction, and full-chain data lineage governance. This directly produces trustworthy token assets and fundamentally breaks down knowledge base silos between multiple Agents.
On the training-inference acceleration front, the AI-in-Lakehouse engine uniformly schedules the KMI training-inference engine, establishing a high-speed direct link from the lakehouse to GPUs, reducing data transfer overhead, increasing GPU throughput, and continuously supporting the incremental iteration of token assets while ensuring high-concurrency, low-latency multi-agent collaborative execution.
AI-in-Lakehouse provides the native intelligent architecture foundation, multimodal capabilities solve token quality, the training-inference engine solves deployment speed, and the Harness engineering system solves scalable delivery — the three working in synergy to truly make the data foundation "run" as a stable, reliable supply engine behind large-scale Agent operations. Full-Stack Synergy:
Connecting the Integrated Chain from Data to Agents
The measure of data's value as a production factor is shifting from "how much data you own" to "how much data is actually used by AI." Tokens are becoming a new benchmark for measuring the depth of data value release. Keen Agentic OS focuses on the core challenges of large-scale Agent deployment, covering the full chain from development, operations, to governance. KeenData Lakehouse, the multimodal lakehouse platform, works from the data foundation side to address the underlying bottleneck of stable, high-quality token supply. The two are deeply synergistic and interoperable, together forming the KeenData Agentic Lakehouse Platform — an AI data foundation that integrates data foundation, multimodal computing, Agent development, and the full training-inference chain. This provides enterprises with an integrated solution for production-grade, large-scale Agent deployment, truly enabling the large-scale implementation of AI technology and the release of business value.
With Data Governance as the Foundation,
Building a Solid Quality Base for Public Resource Trading
The boundaries of AI data foundation capabilities must ultimately be tested in real industry scenarios. Around AI data governance requirements — including data standardization, multi-source data integration, and full lifecycle governance — the KeenData Agentic Lakehouse Platform continues to refine its core capabilities for industries across the board.
Ecosystem Co-Building and Industry Practice Presented Side by Side
KeenData's deployed case studies span core industries including government, energy, and manufacturing, with its capabilities having been validated through engineering practice at over 300 large organizations. On the ecosystem building front, KeenData has completed deep adaptation and interoperability with mainstream domestic computing chips and domestic large models, forming a comprehensive adaptation capability covering "domestic chips, domestic models, domestic data" — committed to advancing the independence, self-reliance, and ecosystem prosperity of China's AI infrastructure. At the same time, KeenData has also explored a token-based flexible pricing model, helping enterprises assess the ROI of large-scale Agent deployment and providing a quantifiable reference for future collaborations.
