Cloudflare Reveals Town Lake Data Platform With AI Analytics Agent Skipper

Cloudflare has made public the architecture behind Town Lake, an internal unified data platform that consolidates access to operational, billing, security, and business information previously scattered across multiple storage and analytics systems. The company processes over a billion events per second globally, generating vast quantities of data that had accumulated in isolated silos—Postgres databases, ClickHouse clusters, Kafka streams, BigQuery datasets, and object storage—making it difficult for teams to discover and analyze information efficiently.

Platform Design and Architecture#

Town Lake operates as a lakehouse system built on open-source and cloud-native technologies: Apache Trino enables federated queries, Apache Iceberg handles data versioning, Cloudflare's own R2 object storage provides the backing layer, and DataHub manages metadata. The key innovation is the ability to join data across different database types—Postgres, ClickHouse, Iceberg tables—in a single query without requiring data movement or replication.

To handle data governance automatically, Cloudflare embedded a default-closed model: newly added datasets start inaccessible until they pass both automated scanning and human review. The company developed an internal tool called Skimmer that combines machine learning classification with rule-based analysis to flag sensitive information. Human reviewers then validate these classifications before anyone gains access to the dataset.

Skipper: Natural Language Analytics#

Atop Town Lake sits Skipper, an AI-powered analytics agent that translates plain English questions into validated SQL queries. Rather than requiring users to write code, Skipper leverages metadata, schema information, transformation lineage, documentation, and runtime inspection to generate accurate results. The system sees heavy use in billing analysis, customer support investigations, business intelligence reporting, and security investigations.

Real-World Usage and Impact#

Billing workloads dominate Town Lake traffic, representing 53% of all queries. In one measurement period, the platform processed over 91,000 billing-related queries from 324 employees alone, covering analysis, support work, and operational reporting. By consolidating what once required manual SQL expertise or labor-intensive investigation, teams can now answer complex questions in seconds.