Show HN: Druids — Build your own software factory

What it is

Druids is an open-source, batteries-included library for coordinating and deploying coding agents across machines. Think of it as orchestration for agentic workflows: the library abstracts VM infrastructure, agent provisioning, and inter-agent communication so you can focus on the logic. The repo demos the pattern with a simple judge-and-workers example — spawn N sandboxed VMs, have each worker submit a PR, then let a judge agent pick the best one. It has been reported that you can run Druids locally or via druids.dev, and allegedly the hosted service supports instant copy‑on‑write clones (agent.fork) for fast experiment iteration.

How it works

Agent programs are just async functions that create agents, register event handlers, and control flow. Events inject deterministic structure — workers trigger "submit", judges trigger "pick", your program decides what happens next. Each agent gets a sandboxed VM with your repo and dependencies; agents can share machines, transfer files, work on git branches, and be messaged while running. The stack is straightforward: a FastAPI server and execution engine, a Python client/CLI, a runtime SDK, and a Vue 3 dashboard. Quickstart requires Docker, uv, and an Anthropic API key — so yes, you’ll need a credit card on file if you want to scale.

Why it matters (and why you should care)

Agent orchestration is the next obvious layer after model APIs. Want to race models against each other, spin up reviewer/implementer pairs, or run large‑scale migrations with automated checks? Druids packages those building blocks. But there’s an emotional tug here: promise and peril, side by side. This could supercharge developer productivity — or, without guardrails, create a lot of noisy, forked branches and runaway automation. DevOps dream or chaos monkey in a suit? Either way, it’s a timely project for anyone thinking beyond single-call LLMs to full‑blown software factories. The code and docs live on GitHub if you want to kick the tires.

Sources: github.com/fulcrumresearch, Hacker News