.NET Global Tool

Ship with a
full AI team

One command scaffolds an autonomous multi-agent AI coding workspace. Orchestrator picks roles, runs the pipeline, delivers PRs. You set direction.

$ dotnet tool install -g multiagent-setup
NuGet version MIT License .NET 10+ 13 providers
multiagent-setup new MyProject — terminal demo

Everything wired up for autonomous AI coding

From install to first autonomous PR in minutes. No YAML, no plumbing, no boilerplate.

🤖
Autonomous pipeline

Orchestrator breaks tasks into steps, assigns specialist roles, and gates each stage on APPROVED / NEEDS WORK. Humans steer direction only.

🔌
13 AI providers

Claude, Gemini, Codex, Qwen, Cursor, Windsurf, Copilot, Nessy, Cline, Aider, Continue, Roo Code, Kiro — one command each, or --provider all to scaffold every provider at once.

🪝
Safety hooks

Block rm -rf /, enforce conventional commits, auto-lint on save — built-in, cross-platform, zero config.

🧠
Knowledge graph

AGE graph + O'Brien semantic memory: agents share context and accumulate project knowledge across sessions.

🔄
Role sync

30+ specialist roles from agency-agents, installed to project-local .claude/commands/ — always current.

Zero shell scripts

Logic lives in the .NET binary. .csx wrappers are the only scripts. Works on macOS, Linux, and Windows.

Set up your AI coding workspace in three steps

Install the tool, scaffold a workspace, tell the Orchestrator what to build.

zsh
# 1. Install the tool
$ dotnet tool install -g multiagent-setup
 
# 2. Scaffold a workspace (Claude by default)
$ multiagent-setup new MyProject
 
# Or pick your provider
$ multiagent-setup new MyProject --provider gemini
$ multiagent-setup new MyProject --provider all
 
# Add a provider to an existing workspace
$ multiagent-setup add-provider cursor
$ multiagent-setup add-provider all
 
# Update existing workspace to latest templates
$ multiagent-setup update
inside workspace
# Enter the workspace
$ cd MyProject
 
# Launch your AI of choice
$ claude
> Claude ready
 
# Give the Orchestrator a task
/orchestrator Build me a REST API with auth
 
↳ PLAN → assigning architect…
↳ BUILD → developer coding…
↳ TEST → tester running suite…
↳ SHIP → PR opened ✓

Works with every major AI coding agent

Each provider gets its own scaffold: config files, hooks, role commands — all wired up.

Claude
--provider claude
Gemini
--provider gemini
Codex
--provider codex
Qwen
--provider qwen
Cursor
--provider cursor
Windsurf
--provider windsurf
Copilot
--provider copilot
Nessy
--provider nessy
Cline
--provider cline
Aider
--provider aider
Continue
--provider continue
Roo Code
--provider roo
Kiro
--provider kiro
all providers at once
$ multiagent-setup new MyProject --provider all
Scaffolded 13 provider workspaces ✓

Autonomous AI agent pipeline: PLAN → SHIP

PLAN
BUILD
TEST
VERIFY
SHIP
✓ APPROVED — advance ⚠ NEEDS WORK — retry ↻ Auto-retry 3× → helper → escalate

5 pipeline types: feature, bugfix (skip PLAN), infra, content, spike (PLAN only). Each step gates on APPROVED / NEEDS WORK. Auto-retry 3× → helper agent → CEO escalation.

Common questions

Do I need .NET installed to use multiagent-setup?

On macOS and Linux you can install via Homebrew — no .NET SDK required: brew install Neftedollar/multiagent-template/multiagent-setup. The dotnet global tool route (dotnet tool install -g multiagent-setup) requires .NET 10 SDK. A bootstrap script is also available that installs all prerequisites automatically.

Which AI coding agents are supported?

13 providers: Claude Code, Gemini CLI, Codex, Qwen, Cursor, Windsurf, GitHub Copilot, Cline, Aider, Continue.dev, Roo Code, Nessy, and Amazon Kiro. Use --provider all to scaffold every provider at once, or add providers later with multiagent-setup add-provider <name>.

Can I add multiagent-setup to an existing code repository?

Yes. Use multiagent-setup init inside any existing git repository. It adds workspace files (CLAUDE.md, docs/, .claude/, hooks) without touching your code or creating a new directory.

Is multiagent-setup free and open source?

Yes — MIT license, free for commercial and personal use. Source code is on GitHub.

How does the orchestrator know which agent role to assign?

The orchestrator reads docs/role-capabilities.md — a capability index generated during workspace setup — and selects roles dynamically. It never hardcodes assignments. If no existing role fits a task it can create an ad-hoc role on the fly.