Multi-Agent War Room
A web app that spawns four specialist agents in parallel, streams their responses to the browser in real time via SSE, then feeds their findings into a synthesizer agent that produces a unified decision brief.
Getting started
Install dependencies:
bun install
bun run build:sdk
Set an API key:
export CLINE_API_KEY="cline_..."
Run:
bun dev
Open http://localhost:3456 in your browser, enter a mission, and watch the agents work.
What it does
- You enter a mission in the browser
- The server spawns four
Agentinstances in parallel viaPromise.all:- Architect (system design)
- Security Analyst (audit)
- Pragmatist (product)
- Skeptic (red team)
- Each agent streams
assistant-text-deltaevents to the browser via SSE, rendered in its own card - Once all specialists finish, a synthesizer agent combines their findings into a unified decision brief, also streamed live
Concepts demonstrated
- Running multiple
Agentinstances concurrently withPromise.all - Per-agent
subscribe()for independent event streams - Server-Sent Events (SSE) to stream agent output to a browser
- Agent composition: feeding one agent's output as input to another
- Inline HTML frontend served from the same Node.js server (single file, no build step)
Notes
For a simpler starting point, see quickstart. For custom tools and structured workflows, see code-review-bot.
