Introducing Radar Researcher: An AI tool for exploring Internet data in plain language

Cloudflare Radar, which since 2020 has published open data on global Internet traffic from its network, has introduced Radar Researcher, a beta tool enabling users to ask questions in plain language and receive real-time, interactive charts without navigating APIs or dashboards. The tool aims to lower technical barriers, making Radar’s datasets—including DNS queries, HTTP traffic, and network quality data—accessible to journalists, researchers, and network operators without requiring expertise in data structures or query languages. Radar Researcher is integrated into every Radar page, allowing users to open an AI panel that processes prompts and generates contextual explanations or visualizations based on the exact data and filters currently applied. The tool addresses the challenge of making complex datasets understandable to a diverse user base, from novices to technical experts, by automating data retrieval and presentation while maintaining transparency through traceable tool calls and reasoning steps.

Radar Researcher is designed to streamline workflows for users under time constraints, such as journalists investigating Internet outages, by eliminating the need to manually search through multiple pages or graphs. For example, during reported Internet shutdowns in Iran in early 2026, the tool automatically gathered and analyzed HTTP traffic data, presenting a timeline with annotated charts and outage records, then suggested further comparisons with neighboring countries. The AI assistant captures the exact visualization a user is viewing, analyzes the underlying data, and provides explanations in plain language, ensuring precision by using raw API data rather than estimated values from visual representations. Users can follow the assistant’s reasoning process through expandable traces that detail the model’s tool calls and data queries, reinforcing trust in the results.

Built entirely on Cloudflare’s developer platform, Radar Researcher operates as a stateful Durable Object with its own SQLite database, ensuring chat history and responses persist even if a user leaves the page mid-conversation. The tool leverages Workers AI, running open models like Kimi K2.7, with a fallback chain across three model families to maintain availability if one provider experiences capacity issues. Every interaction routes through AI Gateway for logging, cost tracking, caching, and safety controls, while the assistant queries Radar’s API indirectly via the Cloudflare MCP server, which dynamically discovers and executes the correct endpoints without hard-coded specifications. This architecture allows Radar Researcher to adapt to new datasets automatically, requiring no manual updates to the AI’s prompt or logic as Radar expands its offerings.

A key innovation in Radar Researcher is its approach to data presentation, avoiding the common issue of language models truncating or summarizing numerical data in favor of rich, interactive charts. Instead of embedding raw numbers in prose, the model generates lightweight chart specifications that reference the API data paths, ensuring users receive the same dynamic visualizations available elsewhere on Radar. This design choice preserves data integrity while delivering user-friendly explanations, and the tool’s transparency extends to showing the exact steps taken to retrieve and process data. Radar Researcher exemplifies how AI can enhance data accessibility without sacrificing accuracy or control, offering a model for integrating AI assistants into complex, real-time datasets while maintaining user trust and operational reliability.