Multi-Agent Explorer

A lead agent decomposes a research question, fans subtopics out to parallel explorer subagents with their own tool loops, then synthesizes a cited report with every hop visible in the stream.

ToolsLoading

Opening demo

The demo workspace is loading. This can take a little longer on slow mobile networks.

Agent DemoLoading

A lead agent that delegates research to parallel subagents

A lead agent decomposes a research question into subtopics, fans them out to explorer subagents that run their own tool loops in parallel, then synthesizes the findings — the whole orchestration stays visible in the chat stream.

Readyopenai/gpt-5.6-luna

One lead, many explorers

Ask a research question. The lead agent will decompose it into subtopics, fan them out to parallel explorer subagents, then synthesize a report while you watch the fan-out happen.

Lead → fan-out → synthesize

A hand-rolled orchestration layer on the Vercel AI SDK. One lead agent plans, explorer subagents run their own tool loops in parallel, and the lead merges their findings.

Plan

The lead decomposes the question into 2–4 disjoint subtopics with generateObject.

Fan-out

Each subtopic gets an explorer subagent with isolated context and its own streamText loop.

Synthesize

The lead streams a report built from the explorers' compact findings.

Explorer tools

search_notes

Scored keyword lookup over a deterministic agent-design corpus.

read_note

Reads one note in full so explorers can cite it.

Runtime

v24.21.0

openai/gpt-5.6-luna