Tacnode Inc. today emerged from stealth mode with a new data infrastructure platform designed to enable artificial intelligence agents to reason and act over a shared, up-to-date representation of enterprise context.

The company’s Tacnode Context Lake technology and Semantic Operators feature form what it describes as a “context layer” for agent-based systems. Context Lake is intended to enable autonomous AI agents to share knowledge, update the collective state and make coordinated decisions in real time. Tacnode said the system provides a unified environment for live data ingestion, incremental transformation and low-latency retrieval, with performance engineered to support millions of retrieval requests per second.

Addressing context fragmentation

Tacnode Chief Executive and Chief Architect Xiaowei Jiang said enterprise data infrastructure was designed around human decision cycles, which move at minutes, hours or days, rather than the millisecond-scale loops common to AI agents. As a result, agents often must operate on fragmented and inconsistent data held across databases, streams, feature stores, vector stores and custom memory layers.

“Data lakes are built to make sure humans can do analytics very efficiently, but over the last couple of years, the focus has been on agents,” he said. “Agents use data very differently from humans.”

Context Lake is similar to a data lake but is architected around distinct use cases.  When many agents make decisions simultaneously, their actions may influence one another, Jiang said. “We need to make sure that when lots of agents are acting concurrently, they affect each other in a constructive way,” he said.

Context Lake maintains a continuously updated context representation for all agents to access and modify. Semantic Operators operate against both structured and unstructured data to enable richer contextual reasoning and “a shared, live view of the business as it is now,” Jiang said.

Tacnode built the product from scratch but integrates with open-source standards such as PostgreSQL and Apache Iceberg. “Everyone who knows PostgreSQL can use it, and we interact with Iceberg very well,” Jiang said.

In use cases where enterprises want analytical processing, Jiang said the system can operate alongside Iceberg. “You can view it as a real-time layer before Iceberg,” he said. “Iceberg is great for analytics, but it’s not really designed for real-time retrieval.”

Tacnode guarantees standard transactional isolation levels to ensure consistency and latency. “Context Lake is able to store structured, semi-structured, unstructured and vector data in one place,” Jiang said. “We provide the transactional consistency across all that data. We have the standard transaction, serializable, snapshot and read committed isolation levels.”

Already in production

Tacnode said its platform is already running in production at food delivery company DoorDash Inc., where it supports in-session personalization by reacting to user behavior within hundreds of milliseconds. It said the same process used to require several minutes.

The company also has “a few other customers” in production, Jiang said. The product is available through the Amazon Web Services Inc. Marketplace, complemented by direct sales.

Tacnode is backed by venture funding, but Jiang declined to disclose investors or amounts. “We want to be focused on the product and the problem we are solving,” he said.

Image: SiliconANGLE/Reve