en.Wedoany.com Reported - Cloudflare has released the latest application status of its internal unified data platform, Town Lake, with billing workloads connected to the platform accounting for 53% of all queries. The platform is equipped with an AI-driven analytics agent named Skipper, designed to unify operational, billing, security, and business data previously scattered across different systems.
Cloudflare's global network processes over one billion events per second, covering more than 330 cities across 120 countries. Over time, data has accumulated in Postgres databases, ClickHouse clusters, Kafka streams, BigQuery datasets, and object storage systems, increasing the difficulty of discovery and analysis. Town Lake serves as a unified SQL interface to access these systems while maintaining data governance and access control capabilities.
The platform is based on a lakehouse architecture, utilizing Apache Trino, Apache Iceberg, Cloudflare R2 object storage, and DataHub for metadata management. A single query can join data across Postgres, ClickHouse, and Iceberg tables without moving data between systems. Supporting services handle data ingestion, transformation, access control, and identification of personally identifiable information (PII).
The platform adopts a closed-by-default governance model, where newly connected datasets require automated scanning and manual review before access is granted. Cloudflare uses a service called Skimmer, which combines automated classification with AI analysis to detect sensitive data, followed by human reviewers to verify classification suitability.
Leveraging Town Lake, Cloudflare built Skipper to provide natural language access to enterprise data. This agent utilizes metadata, schema definitions, transformation lineage, documentation, and runtime checks to convert user requests into validated queries, improving accuracy. The tool is used for billing analysis, customer support investigations, business intelligence, and security workflows.
Dmitry Alexeenko, Head of Enterprise Engineering at Cloudflare, noted that the team built the unified data platform Town Lake and the AI data agent Skipper, which can convert plain English questions into insights within seconds.
Tasks that previously required complex SQL or manual investigation can now be completed in seconds through the unified data platform and AI agent. Billing workloads account for the majority of Town Lake usage. During one measurement period, the platform processed 91,760 billing-related queries from 324 employees, covering billing analysis, support investigations, and operational reporting. The company also found that simplifying AI agent prompts improved accuracy, and consolidating overlapping tools reduced erroneous selections. Incorporating SQL transformation logic and data lineage into the agent context further enhanced business semantic understanding.
Patrick Joubert, CEO of Rippletide, commented that after placing an internal AI agent on top of a unified analytics platform, if the agent can reason over operational data, execution must be close to the operational layer. Distributed deterministic checks can enable the agent to act without turning the data platform into an uncontrollable operational layer.
Cloudflare plans to integrate Skipper more deeply into internal chat, ticketing, and development workflows. It will also expand its Transformer pipeline, allowing teams to define curated datasets using SQL and metadata files, which are automatically deployed, monitored, cataloged, and surfaced through DataHub and Skipper. The company also expects to migrate more Town Lake workloads to R2 SQL in the future.
