Stacks/Multi-agent orchestration

A rolled-your-own Node.js multi-agent setup (HN)

pablovarela's production setup is deliberately plain: Node.js, Express endpoints in V8 isolates, a coordinator that sequences or fans out, MongoDB for shared state and for a complete log of every run's input, output, tokens, and latency. No commercial agent framework is used.

Use case

A production multi-agent system rolled by the author on Node.js, with agents as Express endpoints that can be chained or run in parallel

Category

Multi-agent orchestration

Team size

not stated

Orchestrator / framework

Each agent is an Express endpoint running in a V8 isolate. A simple coordinator endpoint chains agents sequentially or fans out in parallel depending on the task. Agents do not share a framework beyond this.

Memory

Agents communicate through a shared MongoDB layer, reading and writing state as JSON documents linked by a pipeline ID.

Evals

not stated

Observability

Every agent run is logged — input, output, token usage, and latency — to a dedicated MongoDB collection. The author calls this simple and says it catches failures fast.

Monthly cost

not stated

Models

not stated

Observability tools

MongoDB (dedicated run-log collection)

Other tools

V8 isolatesMongoDB

Source

https://news.ycombinator.com/item?id=47695256

pablovarela · Apr 7, 2026