Multi-agent orchestration dashboard
Operations tooling for teams running AI agent systems · Interactive prototype with sample data, solo
Updated
Before
Agents ran as a black box. When a run failed, engineers read raw logs across several services.
After
One screen that shows which agent failed in a run and the message trail that led there.
- 1Person:Engineer
Draws a workflow in the builder and sets each agent's role, tools and limits.
- 2Software:Orchestrator
Assigns tasks to worker, search and tool agents.
Software:AgentsPass messages to each other, each stamped with the run's trace id.
- 3Software:Platform
Updates the live map, latency and cost for every agent as the run goes.
- 4Person:Engineer
Opens the failing agent, follows its message trail and fixes the step.
Always on
- A failed agent retries, then escalates to human review.
- Every message is stored with its trace id.
- An agent over its cost cap is paused.
Problem
Several AI agents worked together, passing messages and branching. When a run failed, nobody could see which agent broke or why.
Diagnosis
Logs from each agent say little alone. The failure sits in the hand-offs, so every message needs one trace id from start to end.
Options I weighed
How does an engineer find the agent that broke a run?
Chosen
One trace id per run, shown as a message trail across agents
The failure usually sits in a hand-off, and the trail shows every hand-off in order. Cost: every message must carry the id.
Rejected
A separate log view per agent
Each log looks fine alone, so the broken hand-off stays hidden.
Solution
A mission-control app for agent runs: live agent map, workflow builder, cost and latency charts, and message tracing across a run, driven by sample runs.
- Interface
- Next.js
- React
- Tailwind CSS
- Radix UI
- Recharts
- Motion
- Delivery
- Vitest
- Playwright
- Cloudflare Pages