Databend Labs, the company behind Databend Cloud, today announced the Databend Cloud Agent Trace Solution, an end-to-end cloud data pipeline engineered to process and analyze agent trace data at production scale.
As AI agents take on complex work, a single run can span millions of context tokens, include thousands of tool calls, and run for hours. Trace data combines model inputs and outputs, tool responses, intermediate states, and nested spans in JSON structures that evolve across models, frameworks, and applications. Logging pipelines often struggle to preserve this data for investigation, evaluation, and historical analysis.
The solution consolidates the trace data lifecycle on Databend Cloud. Raw trace and span events arrive from Kafka and object storage. Databend’s native VARIANT type preserves deeply nested, heterogeneous JSON, allowing teams to retain raw payloads and define field extraction, transformation, and analytical models in SQL. Streams and Tasks process only new events, producing analysis-ready tables for path reconstruction, root-cause analysis, Evals, replay, and training feedback.
Built as an S3-native cloud lakehouse, Databend Cloud separates storage and compute. Independent elastic warehouses isolate ingestion, transformation, and interactive analysis. Automatic scaling handles workload changes, auto-suspend reduces idle costs, and masking policies protect sensitive data at query time.
The solution has been validated by a leading foundation model company that uses Databend Cloud to process traces from its flagship trillion-parameter reasoning model and behavioral data from all its applications. The deployment reached production in two weeks and now handles TB-scale Trace writes per hour.
Compared with Langfuse, an application-layer observability platform that uses a predefined trace-and-observation model, ties historical data access to the selected plan, Databend Cloud retains heterogeneous payloads as native VARIANT data in the customer’s own lakehouse. Customers control data retention through their object-storage policies and use SQL to define JSON structures, field extraction logic, and analytical models for different agents and applications, rather than conforming to a vendor-defined data model. Engineering and research teams use these data assets to investigate long-running agent paths, compare model and harness behavior, and feed findings back into development.
Databend Cloud provides a unified data platform for modern applications and AI workloads, combining raw-data retention, incremental transformation, analytics, search, Evals, and training data preparation without requiring a fragmented data stack. The Databend Cloud Agent Trace Solution is available on AWS.
ABOUT Databend Labs
Databend Labs is the company behind Databend, an open-source, cloud-native lakehouse, and Databend Cloud, its fully managed enterprise cloud platform. Built on an object-storage-native architecture, Databend Cloud unifies ingestion, incremental transformation, and analytics in one elastic SQL platform, helping teams build reliable, replayable data pipelines for modern applications and AI workloads without operating a complex data stack. Learn more at databend.com.
View source version on businesswire.com: https://www.businesswire.com/news/home/20260911753396/en/
Media gallery
