AI Product NewsNetwork Intelligence Tools

Cloudflare’s Radar Researcher Turns Internet Data Into Plain‑English Answers and Interactive Charts

Cloudflare has introduced Radar Researcher, an AI assistant that lets you ask plain‑English questions about Internet outages, traffic shifts, and quality. It translates queries into API calls, returns verifiable explanations, and renders interactive charts—cutting time to insight for journalists, researchers, operators, and policy teams.

NexusAI EditorialAug 10, 20262.8K views8 min read

Cloudflare’s Radar Researcher Turns Internet Data Into Plain‑English Answers and Interactive Charts

AI Brief

Cloudflare’s Radar Researcher moves Internet measurement from dashboards and APIs to conversational analysis. Ask about outages, traffic anomalies, or quality metrics and get API‑sourced answers with interactive charts, audit trails, and suggested follow‑ups. Built on Workers, Durable Objects, and Workers AI with a fallback chain of models and AI Gateway, it emphasizes reliability and verifiability while lowering the barrier for non‑experts. For practitioners, the draw is workflow speed: less time wrangling filters, more time investigating incidents, comparing regions, and embedding evidence in reports. For leaders, it signals how agents will front‑end complex data platforms, shifting value from UI navigation to explainable, defensible answers.

Radar Researcher is Cloudflare’s new conversational front end for the company’s open Internet measurements. Instead of spelunking through Radar pages or hand‑rolling API queries, users can ask plain‑English questions about traffic spikes, outage timelines, or network quality and get responsive, interactive charts plus a short, sourced explanation. The payoff is practical: analysts, journalists, and operators move from hunting the right view to interrogating the data directly, with follow‑up prompts that keep the investigation flowing.

What differentiates the tool isn’t only the chat veneer. Each answer is backed by Radar’s APIs and arrives with the exact visualization a user could have built manually, alongside a trace of how the model interpreted the question. That mix of narrative, chart, and audit trail helps close the trust gap that dogs many AI assistants, making the output citeable in incident write‑ups, policy briefs, or executive updates without resorting to screenshots of third‑party tools.

Under the hood, Researcher runs on Cloudflare’s developer stack: a Worker orchestrates the conversation, Durable Objects persist state, and Workers AI provides model inference with an ordered fallback across model families. AI Gateway handles logging, cost control, and caching. The result is a resilient agent that keeps generating even if a tab closes, and returns the same structured datasets you would fetch through Radar directly—only faster and with context applied to your current view.

For buyers and builders, the strategic signal is clear: observability platforms are converging on agentic interfaces that reduce time‑to‑evidence. Expect teams to standardize on conversational investigation for outage triage, regional comparisons, and media‑ready summaries. The risk shifts to governance—ensuring claims map to first‑party data and that prompts, query plans, and visualizations are reviewable. Radar Researcher’s explicit audit path and API‑sourced charts are a strong template for that future.

Key Takeaways

Conversational Analysis Cuts Time to Evidence

Plain‑English prompts now fetch Radar datasets and render interactive, shareable charts—reducing the effort to move from question to defensible insight.

Trust via API‑Backed Charts and Audit Trails

Each answer includes how the model interpreted the query, which datasets were queried, and official Radar visualizations to substantiate claims.

Clear Path to Embedding and Governance

Running on Workers, Agents SDK, and AI Gateway makes it extensible; teams can add prompts, guardrails, and export flows while enforcing review policies.

What Changed: Ask Radar in Plain English

Researcher adds a conversational layer to Cloudflare Radar, letting users pose questions such as “Show the outage window and recovery timeline for Country X last quarter” or “Compare broadband latency for Portugal and Spain over six months.” The assistant translates the request into Radar API calls, returns the same interactive visualizations on Radar, and includes a short explanation plus suggested follow‑ups. It also supports “Explain with AI” from any chart, seeding the chat with your exact data, filters, and date range.

Under the Hood: Architecture Built for Trust

A Cloudflare Worker manages conversations; each session persists in a Durable Object with SQLite so history and streaming responses survive disconnects. Workers AI powers the LLM, with a prioritized fallback chain across model families to avoid capacity stalls. AI Gateway provides logging, cost tracking, and caching. Crucially, the agent passes the model both the raw dataset and your view parameters, then renders official Radar charts—so numbers are API‑grounded, explanations reference first‑party data, and reasoning steps are auditable.

Who Benefits and How to Use It

• Journalists: Build outage timelines and regional comparisons in minutes, then share a link to the conversation for editorial review. • Network teams: Kick off incident triage with a natural‑language probe that auto‑pulls the right metrics and annotations. • Researchers and policy staff: Generate reproducible analyses with verifiable charts to back claims. Practical flow: ask a broad question, inspect the audit trail, iterate with targeted follow‑ups, and export the charts or share the conversation for peer validation.

Limits, Risks, and Governance

No agent removes the need for judgment. Explanations may over‑fit to short time windows, conflate correlation with causation, or miss edge cases when datasets are sparse. Teams should treat the audit view and underlying API responses as the source of truth, standardize prompts for recurring analyses, and require human review for claims likely to reach external audiences. When comparing countries, align date ranges, normalization choices, and sampling windows to avoid misleading narratives.

Evaluation Checklist and Integration Path

Pilot with three tasks: 1) Explain a known outage and verify timing versus past methods; 2) Compare network quality across two regions with identical filters; 3) Produce a one‑page brief with annotated charts. Assess speed, reproducibility, and clarity of the audit trail. For deeper integration, use Workers and the Agents SDK to add house prompts, guardrails, and export targets (ticketing, newsroom templates, Slack). Establish a policy that any externally shared figure must include the Researcher chart link or the direct Radar API query.

Frequently Asked Questions

How do I validate an answer before publishing?

Open the audit view to see the parsed intent, datasets, and parameters; cross‑check the numbers in the returned interactive chart; then copy the request as a Radar API call for independent verification. Lock date ranges, normalization choices, and filters, and include the chart or API reference in your doc.

What kinds of questions does Radar Researcher handle well?

It excels at outage timelines, regional traffic comparisons, and Internet quality metrics (latency, throughput, consistency). It can start from a blank prompt or analyze any existing Radar chart via “Explain with AI.” For causal claims, pair its narrative with outside context and treat the API‑sourced chart as the factual backbone.

Can my team extend or embed it in our workflows?

Yes. Because it’s built on Workers and the Agents SDK, you can add house prompts, role‑based guardrails, and export actions (e.g., ticketing, newsroom briefs, Slack). Use AI Gateway for cost control and logging, and standardize prompts for recurring analyses to ensure consistent outputs.