Chorus

Hover to perturb the swarm

The
Lab

02 / Lab

Chorus Labs is an applied AI research lab. We believe intelligence isn't one giant model. It is hundreds of specialised agents colliding, challenging each other and settling on answers you can trust.

Every sphere above is an agent. Swarm, our platform, lets teams spin up, coordinate and evaluate agent collectives in production, with the rigour of a research lab and the reliability of infrastructure.

100+Agents per run
40msMedian routing latency
12Papers published

The Swarm
Platform

Six products, one idea: many small models working together beat one big model working alone.

Platform No. 01✦

Swarm
Orchestrator

Coordinate 100+ agents on one goal

Generally available
ML-01-2026
Research No. 02✦

Adaptive
Memory

Context that reshapes itself

Public beta
ML-02-2026
Tooling No. 03✦

Eval
Studio

Benchmark every agent, every run

Generally available
ML-03-2026
Data No. 04✦

Synthetic
Bloom

Grounded synthetic training data

Public beta
ML-04-2026
Infra No. 05✦

Gravity
Router

Send each task to the right model

Generally available
ML-05-2026
Safety No. 06✦

Consensus
Layer

Agents cross-check before they act

Research preview
ML-06-2026

How Swarm
Works

04 / Platform

Connect

Plug in your data, tools and existing models through one API. Swarm maps what each agent is allowed to touch.

  • API
  • Connectors

Orchestrate

Describe the goal. Swarm spawns specialist agents, routes sub-tasks and lets them debate until the answer settles.

  • Routing
  • Debate

Evaluate

Every run is scored for accuracy, cost and safety, so regressions surface before your users ever see them.

  • Evals
  • Tracing

Deploy

Ship to our cloud or your own VPC with audit logs, SSO and per-agent permissions built in.

  • VPC
  • SOC 2

The Model
Family

Six in-house models, each tuned for a role in the swarm. Hover one to open it up.

Edge No. 01

Mote

Runs on laptops and phones. Filters and routes work before it ever reaches a bigger model.

1B · On-device

Classification & triage
ML-M01
Fast No. 02

Petal

First drafts, summaries and tool calls at a fraction of the cost of larger models.

8B · 128K context

Fast drafting agents
ML-M02
Core No. 03

Core

The default agent in most swarms, balancing speed, cost and quality.

70B · 256K context

General-purpose workhorse
ML-M03
Flagship No. 04

Aurum

Plans the run, breaks goals into tasks and arbitrates when agents disagree.

Flagship · 1M context

Deep reasoning & planning
ML-M04
Realtime No. 05

Pulse

Low-latency speech in and out, for agents that talk directly to people.

Realtime · <200ms

Voice & streaming
ML-M05
Embeddings No. 06

Ink

Powers Adaptive Memory and retrieval across every agent in the swarm.

Embeddings · 3072-d

Search & memory
ML-M06
“Intelligence isn't one big model. It's many small ones colliding until the right answer settles.” Chorus Labs Research Notes
06 / Early Access Research preview open
EA
Early Access No. 001✦

Let your
agents collide

Swarm is onboarding a small group of research and product teams each month. Tell us what you're building and we'll get you set up with credits and a lab engineer.

Admit One ✦

Swarm v1.0

Research
Preview

ML-EA-2026