research · tool agent
Web Research Agent
Turn an open-ended question into a sourced brief by observing the web, not by improvising citations.
- research
- search
- browser
- verification
- tool-calling
How to watch
Play starts at Input. Each station works, then a labeled packet travels the wire.
On station · 5.2s · 0/0
Packets stay on the wire. This is a conceptual simulation, not a live model run.
What Web Research Agent does
A web research agent is an AI system designed to gather, evaluate, and synthesize information from web sources to complete a research task. It does not answer from memory alone. It searches, opens pages, extracts evidence, and checks whether the evidence actually supports the claim.
The agent receives a research question, writes a short plan, calls a search tool, opens a subset of results in a browser or fetch tool, extracts quotations and facts, compares them, and only then writes an answer. If search fails or sources conflict, it retries or states the uncertainty. This page shows a conceptual simulation of that observable loop — not a model’s private chain-of-thought.
Workflow
Each step maps to a node in the lab above. and watch the packets travel.
- 01Input
Receive question
The agent accepts the research question and any limits such as recency or region.
Data preview
In
What changed in Next.js App Router metadata APIs in 2025–2026?
Out
Normalized task with scope: official docs first, cite URLs.
- 02Plan
Plan the investigation
The agent writes a short plan: official docs, then secondary explainers, then a recency check.
Data preview