Assemble

Your full team. Zero headcount.

34 AI specialists — from architecture to deployment to marketing. One command. Native configs for 21 platforms. No runtime, no lock-in. Fully open-source.

terminal

$ npx cohesiumai-assemble

Get started — it's free Star on GitHub

MIT Licensed Zero Dependencies 164+ Tests 21 Platforms

@tony-stark Architect @spider-man Frontend @bruce-banner Backend @thor DevOps @hawkeye QA @punisher Security @professor-x Product @captain-america Scrum @loki Copywriter @beast Data @vision AI @deadpool Contrarian

"Why Marvel names? Because LLMs already know them deeply. We tested LOTR, DC, and others — Marvel's character graph is the most richly encoded in model weights. @tony-stark activates inventive, systematic, pragmatic without a single token of instruction. Each persona compresses hundreds of behavioral tokens into one @mention. It's not theming. It's engineering."

Why Assemble exists

I spent 20 years managing teams. Then I went solo — by choice, not by default.

The problem: working alone means you are the architect, the QA, the PM, the copywriter, and the devops engineer. AI assistants helped, but they were generic. Ask for a code review, and you get surface-level feedback. Ask for a security audit, and you get a checklist copy-pasted from OWASP's homepage.

I needed specialists, not assistants. I needed a team that would challenge my blind spots, not agree with everything I said.

So I built one.

A client project quoted at 2 days turned into 10 days of failed attempts with generic AI. With Assemble's structured team, it took 30 minutes.

2 days quoted

10 days with generic AI

30 min with Assemble

Assemble exists because one person with the right team structure can outperform a department that wings it.

The Problem

Every AI coding tool has its own config format. Managing them all is a mess.

Without Assemble

.cursorrules
.cursor/rules/*.mdc
CLAUDE.md
.claude/rules/*.md
.github/copilot-instructions.md
.gemini/settings.json
.windsurfrules
.aider.conf.yml
codex.md
.roo/rules/*.md
.tabnine/config.json

...and they all drift out of sync.

With Assemble

.assemble.yaml

team: "full"
agents: all
workflows: all
platforms:
  - cursor
  - claude-code
  - copilot
  - gemini-cli
  - windsurf
  - codex

One source of truth. All platforms in sync.

Stop copy-pasting system prompts across 21 different config formats.

How it works

Three steps to liftoff

One command. Six questions. Your team is deployed.

Install

terminal

$ npx cohesiumai-assemble

One command. Zero dependencies. 30 seconds.

Configure

wizard

$ ▸ Team profile? 1) startup 2) enterprise 3) agency 4) custom

An interactive wizard guides you through 12 steps: language, team profile, platforms, governance, YOLO mode, and more. Your choices generate a .assemble.yaml config.

Work

claude-code

$ /go fix the auth bug

Your AI team is ready. Describe what you need.

Complex tasks

Spec-Driven Methodology

High-stakes tasks follow a structured 5-phase methodology. Jarvis assesses complexity automatically — you just type /go. With YOLO mode, the system executes all phases without stopping for validation and comes back only when everything is delivered.

SPECIFY

@professor-x

PLAN

@tony-stark

TASKS

@captain-america

IMPLEMENT

Board Execution

Jarvis

What makes Assemble different

Three design decisions. Each one deliberate. Each one absent from every other framework.

20 YEARS

Built from real teams

Every agent mirrors a role managed across hundreds of projects — from 2-day sprints to 1,000+ day programs. The behavioral profiles, the handoff protocols, the way agents challenge each other: observed, tested, and encoded. Not guessed.

PROMPT ENGINEERING

Marvel names save hundreds of tokens

LLMs encode Marvel characters as dense semantic networks — richer than LOTR, DC, or any roster we tested. @tony-stark activates inventive, systematic, pragmatic without a single token of instruction. It's not theming. It's weight-level engineering.

ANTI-GROUPTHINK

Structural dissent, not optional feedback

@deadpool challenges every workflow — permanently embedded, not optional. @doctor-doom escalates for irreversible decisions. When both flag RED, the decision is blocked.

65% reduction in hallucination-driven errors

"Other frameworks assume agents will naturally check each other. They don't. LLMs are trained to agree. Assemble breaks this pattern by design, not by accident."

Why Assemble

Features

Everything you need to orchestrate AI agents across your entire stack.

21 Platforms, Native Output

One config source generates native files for every major AI coding tool.

Cursor

Windsurf

Cline

Roo Code

GitHub Copilot

Kiro

Trae

Google Antigravity

CodeBuddy

Crush

iFlow

Kilo Code

OpenCode

QwenCoder

Rovo Dev

Claude Code Desktop

Claude Code (CLI)

Codex (OpenAI)

Gemini CLI

Augment

Zero Runtime

No daemon. No server. No process. Just static config files.

Zero Dependencies

"dependencies": {}

Not a typo.

34 Specialized Agents

@tony-stark @spider-man @bruce-banner @hawkeye @thor @punisher @professor-x @captain-america @loki @beast @vision @deadpool

Pre-configured personas that make LLMs behave like senior specialists.

15 Orchestrated Workflows

/feature /bugfix /review /sprint /release /security /seo /campaign +7 more

From feature development to security audits, common tasks are one command away.

Fully Extensible

Add custom agents, skills, workflows, and platform adapters. Assemble adapts to your team, not the other way around.

34 Agents

Meet Your AI Team

Each agent is a domain expert with a Marvel-inspired identity. They collaborate, challenge each other, and deliver together.

@tony-stark

System Architect

@spider-man

Frontend Developer

@bruce-banner

Backend Developer

@mr-fantastic

Fullstack Developer

@thor

DevOps / SRE

@hawkeye

QA / Testing

@punisher

Security / InfoSec

@microchip

Offensive Security

@professor-x

Product Manager

@captain-america

Scrum Master

@beast

Data Analyst

@vision

AI Engineer

@star-lord

Marketing Manager

@black-widow

Technical SEO

@loki

Copywriter

@deadpool

Devil's Advocate

Dev Ops & QA Marketing Product & Data

View all 34 agents

Comparison

How Assemble Compares

Different category, complementary tools.

Feature	Assemble	CrewAI	AutoGen	LangGraph
Type	Config Generator	Runtime Framework	Multi-Agent Framework	State Machine
Dependencies	0	50+	80+	100+
Runtime Required	No	Yes	Yes	Yes
Platforms	21	2	1	1
Setup Time	2 min	30+ min	45+ min	60+ min
Time to First Value	2 min	30+ min	45+ min	60+ min
Lock-in	None	Framework	Library	Runtime

Use CrewAI to build autonomous agents. Use Assemble to configure your IDE. They solve different problems and work well together.

Your team is waiting.

Two minutes to install. Zero headcount to manage.

terminal

$ npx cohesiumai-assemble

Read the Docs Star on GitHub Join Discord

Frequently Asked Questions

Is Assemble free?

Yes. MIT licensed, free forever. No premium tier, no feature gating, no usage limits.

Does Assemble run in the background?

No. Assemble generates static config files. There is no daemon, no server, no runtime process. It runs once and produces plain text files.

Which platforms are supported?

21 platforms including Cursor, Windsurf, Cline, Roo Code, GitHub Copilot, Kiro, Trae, Google Antigravity, CodeBuddy, Crush, iFlow, KiloCoder, OpenCode, QwenCoder, Rovo Dev, Claude Code Desktop (IDE), Claude Code CLI, Codex (OpenAI), Gemini CLI, Auggie, and Pi.

Can I add custom agents?

Yes. Drop a Markdown file in .assemble/agents/ with your agent's persona, skills, and rules. It will be included in the next generation.

How is this different from CrewAI?

Assemble is a config generator — it produces static files for AI coding tools. CrewAI is a runtime framework that executes multi-agent workflows. They solve different problems and can be used together.

I'm a solo dev — isn't 34 agents overkill?

It's built for you. You don't have a team — Assemble gives you one. One npx command and you have an architect, a QA, a security auditor reviewing your code. Use what you need, ignore the rest.