LF Edge TAC session. How EdgeLake exposes industrial edge data without moving it, no cloud upload, no database to manage, and what MCP-driven AI can do directly against the live data.
At the LF Edge TAC meeting on February 11, 2026, the EdgeLake team presented how the project (now a stage-2 LF Edge initiative) lets industrial data stay where it is generated while remaining queryable as if it were centralized. The session also previewed work going to the ProveIt industrial trade show the following week.
The core architecture argument: instead of moving raw PLC and sensor data to a central historian, EdgeLake creates a single query interface that knows where every node lives. A SQL request lands once, EdgeLake distributes the relevant sub-queries to the nodes holding the data, and only the result sets traverse the network. The raw data stays at the edge.
The live demo showed drill-down via the UNS asset framework: enterprise to site to process to subprocess to tank to measurement. The tables backing each asset are generated automatically by EdgeLake when the southbound source is connected, so the operator never thinks about which database holds lot number 16. Three sites each collect only their own data, and the namespace stitches them together into a historian-like view that happens to be fully distributed.
The team also touched on the AnotherPeak boat-fleet deployment as a real-world example. When 4G bandwidth from a boat in the middle of an Alpine lake is not enough to stream raw telemetry, the engine runs locally and pushes only decision outputs to the operator. The TAC discussion closed on monitoring network equipment, where petabyte-scale generation and the need for both real-time access and long-term centralized trends is exactly the split EdgeLake is built for.