19 Agents. One System. Ship Faster.

Nimbus Studio is a multi-agent developer system. An orchestrator dispatches work to 19 specialized AI agents that collaborate to build, test, and deploy software — autonomously.

See How It Works

How It Works

From a single request to a fully deployed application, in three steps.

Describe

Tell the orchestrator what you want to build. A product brief, a feature request, a bug to fix.

bash

curl -X POST http://localhost:18785/agents/dispatch \
  -H "Authorization: Bearer $TOKEN" \
  -d '{
    "agent": "product-owner",
    "task": "Build a showcase website",
    "project": "devnimbus"
  }'

Orchestrate

The orchestrator analyzes your request, selects the right agents, and creates a coordinated execution plan.

Request

product-owner

planner

task-manager

execution queue

Ship

Agents build in parallel. Frontend, API, database, infrastructure — all working simultaneously with dependency-aware scheduling.

frontend

api

database

infrastructure

testing

0sparallel executiondone

19 Agents, 5 Clusters

Every agent is a specialist. Grouped by function, dispatched by need.

Planning

Define what to build, how to build it, and who does what

Building

Write code, design interfaces, and deploy infrastructure

Quality

Test, verify, and polish every deliverable

Content

Create copy, media assets, and documentation

Operations

Monitor system health and manage credentials

Dispatch Architecture

One API. Any agent. Sync or async.

Direct synchronous call to a single agent.

bash

curl -X POST http://localhost:18785/agents/dispatch \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "agent": "frontend",
    "task": "Build the hero section with animated node graph",
    "project": "devnimbus",
    "topic": "hero"
  }'

Response

json

{
  "agent": "frontend",
  "status": "completed",
  "output": "Hero section built with animated SVG node graph, framer-motion entrance animations, and responsive layout. 3 files created."
}

The Message Bus

Agents coordinate through a shared message bus. Channels, inboxes, and queues — all over HTTP.

#orchestrator

Step 1 dispatched to frontend|Step 2 dispatched to database

#status

frontend: building hero section...|database: migrations complete

#errors

No errors

#sprint-1

Database schema ready for review

All communication over HTTP — any language, any tool, any agent

Post a message

curl -X POST http://localhost:18785/bus/channels/status/messages \
  -H "Authorization: Bearer $TOKEN" \
  -d '{"content": "Hero section complete", "agent": "frontend"}'

Read a channel

curl -s -H "Authorization: Bearer $TOKEN" \
  "http://localhost:18785/bus/channels/status/messages"

Response:

[{"agent":"frontend","content":"Hero section complete","timestamp":"2026-03-09T12:34:56Z"}]

MCP Toolkit

Every agent gets superpowers. Model Context Protocol servers extend capabilities beyond code generation.

Voice

voice-tts

Speech synthesis and transcription. Agents can speak and listen — enabling voice-driven development workflows.

Text-to-speechSpeech-to-textVoice profilesMulti-language

Browser

playwright

Full browser automation. Agents navigate, click, type, screenshot, and test web applications in real browsers.

Page navigationElement interactionScreenshotsForm filling

Components

shadcn

Access to the shadcn/ui component registry. Search, view, and install production-quality React components.

Component searchCode examplesInstallationRegistry browsing

Styles

tailwindcss

Tailwind CSS utilities, documentation, and configuration. Convert CSS to utilities, generate color palettes.

Utility searchCSS conversionColor palettesConfig guides

Docs

context7

Live documentation retrieval for any library. Query up-to-date docs and code examples in real-time.

Library lookupCode examplesAPI referenceVersion-specific docs

3D

three.js

3D model processing and React Three Fiber component generation. Convert GLTF/GLB models to JSX components.

Model structureJSX generationMesh optimizationInstancing

Multi-Runtime

No vendor lock-in. Each agent can run on any AI model. Choose the right runtime for each task.

Switch models per-task. Opus for architecture, Haiku for routing, Sonnet for everything between.

runtime-selection.sh

curl -X POST http://localhost:18785/agents/dispatch \
  -d '{
    "agent": "frontend",
    "task": "Build component",
    "model": "opus"
  }'

Team Dispatch

Orchestrate multi-agent builds with dependency-aware execution.

Plan

Build

Verify

Document

product-owner

planner

task-manager

frontend

database

infrastructure

testing

cleanup-qa

copywriter

documentation

PlanBuildVerifyDocument

create-queue.sh

# Create execution queue
curl -X POST http://localhost:18785/bus/queue \
  -d '{"project":"devnimbus","steps":[...]}'

queue-status.json

# Check queue status
curl -s http://localhost:18785/bus/queue/q_abc123

{
  "queue_id": "q_abc123",
  "total": 8,
  "completed": 5,
  "running": 2,
  "pending": 1
}

Live Metrics

Real-time visibility into every agent, every message, every dispatch.

Agent Presence

orchestrator

active for 12m

frontend

active for 8m

database

active for 6m

product-owner

idle

planner

idle

task-manager

idle

infrastructure

idle

testing

idle

copywriter

idle

observer

idle

Bus Activity

Messages/min

Active channels

Queue depth

Activity (last 20 min)

Dispatch Stats

Total dispatches

Success rate

0.0%

Avg duration

Active sessions

Queue utilization78%

presence.sh

curl -s -H "Authorization: Bearer $TOKEN" \
  http://localhost:18785/agents/presence