Agent Swarmdocs

01 -Overview

Many AI agents on the problem. One answer back.

Agent Swarm runs several agents on the same question in parallel, each in a different role, and merges their answers into one deterministic synthesis. Start with a bundled swarm - say, for reviewing a diff or getting customer-style feedback on a launch - or build your own.

Every swarm: pick agents run rounds resolve one synthesis

Run one in a line

Start with a bundled swarm - no configuration required:

# install
npm install -g @calvinnwq/agent-swarm

# run a bundled swarm against a question (2 = rounds)
agent-swarm run 2 "Should we ship the freemium tier this quarter?" \
  --preset customer-panel

Every agent runs at the same time, so a four-agent panel takes about as long as one. You get back a single synthesis you can read, review, or check in.

Ready-made swarms

Each bundled preset has its own roster of agents and its own decision vocabulary. Pick one, or use it as a starting point for your own.

What you get back

The synthesis collects one recommendation in the swarm's own decision vocabulary, how much the panel agreed, the risks they share, and what's still open. No model decides the final word, so the same agent notes always produce the same synthesis.

Synthesiscustomer-panel · round 2 · 3 agents

Fix now - onboarding hides the one action that shows value.

consensus2 of 3 agree
confidencehigh
Still openDoes the empty state need a sample project, or just a clearer first step?

Where next

Note

The README is the authoritative user-facing contract; this site mirrors it. For the durable spec, see SPEC.md.