Cloudflare Ships Skipper AI Agent and Town Lake Data Platform

On May 28, 2026, Cloudflare announced the internal launch of Town Lake, a unified data analytics platform, and Skipper, an AI-driven data agent built to operate on top of it. The combined system consolidates Cloudflare’s internal data operations, which process over one billion events per second, into a single interface. Town Lake removes the need for engineers and analysts to manage separate credentials and query languages across independent systems like Postgres, ClickHouse, Kafka, BigQuery, and R2.

Town Lake Infrastructure

The Town Lake platform is constructed entirely on Cloudflare’s native infrastructure stack. It relies on Cloudflare R2 for long-term storage and Cloudflare Workers for compute and query execution. Cloudflare Workflows handles complex data orchestration tasks, while Cloudflare Access governs authentication and provides automated detection of Personally Identifiable Information.

By centralizing these storage and compute resources, Town Lake provides a unified SQL interface for all company data. This includes metadata, raw network logs, and high-level analytics. The system natively supports distinct handling pipelines for different operational needs. It offers “fast” downsampled data streams used primarily for rapid dashboard rendering, alongside “accurate” unsampled data reserved for critical operations like billing pipelines and deep security investigations.

Skipper Data Agent

Skipper sits directly on top of Town Lake to democratize data access for employees across the organization. Users submit plain English questions to the agent, which translates the request into auditable SQL queries and returns mathematically precise answers in seconds. While the system currently routes requests through Anthropic Claude models via Cloudflare’s integration with Claude Managed Agents, the architecture treats underlying models as commodity APIs to remain model agnostic.

Unlike standard text generators, Skipper incorporates a robust structural validation layer designed to make the AI output readable and maintainable. The agent relies on a sound, incremental TypeScript implementation to perform strict reachability analysis and call-graph mapping. This mapping ensures that any generated tool calls are safe and syntactically correct before the SQL executes against the Town Lake backend. Teams building similar natural language interfaces frequently rely on structured output to enforce type safety between the model and the database driver.

Internal Adoption and Scale

The rollout of Town Lake and Skipper follows the infrastructure expansions announced during Cloudflare Agents Week in April and May of 2026. The new platform leverages recently released tools like Cloudflare Mesh, which grants autonomous agents secure access to isolated private databases. The execution layer relies heavily on Dynamic Workers, an isolate-based runtime specifically designed to execute AI-generated code snippets in milliseconds.

Internal adoption metrics indicate substantial reliance on the new generative stack across the company. As of May 2026, Cloudflare’s internal AI infrastructure serves 3,683 internal users. The underlying systems have already processed 241 billion tokens through the Cloudflare AI Gateway. If you are configuring internal agents for large engineering organizations, testing and evaluating AI agents at this scale requires strict monitoring of query latency and validation failure rates.

Consolidating disparate data silos behind a single AI-accessible interface drastically reduces the time analysts spend hunting for credentials and writing specialized syntax. When building your own internal data agents, abstract your storage backend behind a unified SQL layer and strictly enforce query validation before execution to prevent rogue AI queries from impacting production data stores.