Cloudflare Radar Researcher Lets You Query Internet Data in Plain Language
Quick Reads
- Cloudflare has launched Radar Researcher in beta as part of its Agents Week.
- The tool lets users ask questions about global Internet traffic and get real, interactive charts back.
- It runs entirely on Cloudflare’s developer platform, Workers AI, Durable Objects, and AI Gateway.
- No API knowledge or filter setup is needed. Just type a question in plain language.
- Chat history is saved, shareable via link, and survives mid-session page exits.
Cloudflare has launched Radar Researcher, a new AI-powered tool for exploring global Internet trends and traffic data using plain language. It is live in beta today and available on every page of Cloudflare Radar. cloudflare
The tool solves a real problem. Getting meaningful answers from Radar used to mean navigating pages, setting filters, or writing API queries from scratch. Now, users just ask a question. Cloudflare Radar Researcher handles the rest.
It turns natural language queries into real, interactive charts, built entirely on Cloudflare’s Developer Platform. The panel docks beside whatever the user is already browsing. Research flows without interrupting the existing workflow.
Since launching in 2020, Cloudflare Radar has provided unique insight into global Internet traffic. Its open data serves journalists, researchers, human rights advocates, and network operators. Yet as the dataset grew, so did the complexity of using it. Radar Researcher is the direct answer to that.
Responses come with the same interactive charts found across the rest of Radar, not raw numbers. Each answer includes a plain-language explanation and a real chart. Conversations are saved, searchable, and shareable via a link. Links expire automatically after 30 days. Users can also view a full trace of how the model reached each answer, including every dataset it searched and every API call it made.
One standout feature is Explain with AI. Wherever a chart appears on Radar, one click opens Cloudflare Radar Researcher with that visualization already loaded. The tool captures a screenshot of the chart, pulls the raw data from Radar’s API, and reads the current view parameters, location, date range, and filters. The result is a precise, contextual explanation tied to exactly what the user is looking at. Nothing generic. Nothing estimated.
The tool also handles open-ended investigations well. When asked about Iran’s Internet traffic during early 2026’s government-directed shutdowns, Radar Researcher pulled the outage events Cloudflare Radar recorded for Iran, then gathered the matching traffic data. It produced a clear timeline in seconds. HTTP traffic collapsed from around 0.58 on January 7 to near zero by January 9. Recovery began around January 17. By January 27, traffic had returned close to pre-shutdown levels. That kind of multi-dataset lookup would take a journalist significant manual effort to replicate. Cloudflare covered the event in a dedicated post: Iran’s Internet partial restoration in May 2026.
Under the hood, Cloudflare Radar Researcher runs on a Cloudflare Worker using the Cloudflare Agents SDK. Each conversation lives in a stateful Durable Object with its own SQLite database. Chat history, titles, and streaming responses persist, even if the user leaves the page mid-answer. Generation continues server-side and resumes on reconnect.
The AI engine runs on Workers AI with an ordered model fallback chain. If one model is at capacity, the request cascades to the next. The tool stays available even when a single provider has an incident. All calls route through AI Gateway for logging, caching, cost tracking, and safety guardrails.
Instead of hard-coding tool calls for each of Radar’s hundreds of endpoints, the team connected Radar Researcher to the Cloudflare MCP server via Code Mode. The model gets just three tools, search, execute, and docs. It then writes code to query Radar’s API directly. As Radar adds new datasets, Researcher picks them up automatically. No code changes required.
Charts render through a lightweight specification that references the API path directly. Numbers stay precise because raw data never passes through the model’s prose. The same visual components used across Radar power these charts, time series, stacked areas, bar charts, maps, and histograms.
Cloudflare has also added WebMCP support to Radar. This emerging web standard lets browser-based AI agents call well-defined tools on a page directly, rather than scraping the DOM. External agents can now filter Radar by country, date range, or ASN, search entities, and trigger actions like URL scanning, all programmatically. Radar already passes its own agent-readiness check.
