What is IdeaCo?

IdeaCo is not another multi-agent framework. It's an AI employee management system — you run a virtual company, hire AI agents as employees, assign tasks, and they collaborate autonomously to produce real deliverables.

Each employee has persistent memory, a unique personality, social relationships with coworkers, and can be powered by different backends — from cloud LLMs to local CLI tools like Claude Code, Codex, and CodeBuddy.

The name comes from Idea Unlimited Company (金点子无限公司), inspired by Yang Hongying's children's story, playfully turning a "Limited Company" into "Unlimited".

🚀 Quick Start

Option 1: NPM (Recommended)

Requires Node.js 20+

# Install globally
npm install -g ideaco

# Start your company (opens dashboard automatically)
ideaco start

# Open the dashboard
ideaco ui

Option 2: Desktop Client (Mac/Windows/Linux)

Grab the latest release for your platform:

Download IdeaCo

| Platform | File | |----------|------| | macOS (Apple Silicon) | IdeaCo-x.x.x-arm64.dmg | | macOS (Intel) | IdeaCo-x.x.x-x64.dmg | | Windows | IdeaCo-x.x.x-Setup.exe | | Linux | IdeaCo-x.x.x.AppImage / .deb |

Since IdeaCo is not signed with an Apple Developer certificate, macOS will show a security warning. To open it:

1. Right-click the app → Open → Open again in the dialog
2. Or: System Settings → Privacy & Security → scroll down → Open Anyway
3. Or via Terminal: xattr -cr /Applications/IdeaCo.app

Option 3: From Source

git clone https://github.com/ymssx/IdeaCo.git
cd IdeaCo
yarn install
yarn dev

Option 4: Docker

docker compose up -d

Data is persisted via Docker volumes (app-data, app-workspace).

🧬 Why IdeaCo?

Most agent frameworks create stateless workflows. IdeaCo manages long-living AI employees with memory, personality, and social bonds.

| | Typical Agent Framework | IdeaCo | |---|---|---| | Agents | Ephemeral, per-task | Persistent employees with memory & personality | | Memory | None or simple RAG | Layered: long-term + short-term + social impressions + rolling summary | | Social | Agents don't know each other | Employees form opinions, track affinity, and remember coworkers | | Orchestration | DAG / workflow graph | Company org structure — departments, teams, roles | | Backend | Single LLM | Mix LLMs + CLI tools (Claude Code, Codex, CodeBuddy) | | Autonomy | Triggered by code | Employees poll, think, decide to speak or stay silent on their own | | Interface | Code / YAML | Visual — pixel office, group chats, dashboards |

Architecture

┌───────────────────────────────────────────────────┐
│                  👤 User (Boss)                    │
├───────────────────────────────────────────────────┤
│                🧑‍💼 Secretary                       │
│     Intent Parsing · HR · Task Dispatch · Reports │
├───────────────────────────────────────────────────┤
│               🏢 Organization                      │
│       Company · Department · Team · Requirement   │
├───────────────────────────────────────────────────┤
│                👥 Employee                          │
│     Memory · Personality · Skills · Lifecycle     │
├───────────────────────────────────────────────────┤
│                🤖 Agent                             │
│       LLM Agent · CLI Agent · Web Agent           │
│      (Unified interface, zero business logic)     │
└───────────────────────────────────────────────────┘

• Agent — Pure LLM communication engine. Supports API, CLI, and Web backends behind a single chat() interface. Zero business logic.
• Employee — The atomic unit. Wraps an Agent with persistent memory, personality, skills, and autonomous poll-think-reply behavior.
• Organization — Company structure: departments, teams, group chats, requirements workflow.
• Secretary — Boss's AI assistant that understands intent, manages HR, and coordinates departments.

📖 See ARCHITECTURE.md for the full deep-dive — employee lifecycle, memory system design, flow-of-thought pipeline, anti-spam gates, and social memory details.

⚙ Configuration

LLM Providers

Configure via the Setup Wizard on first launch, or later through the Brain Providers page.

| Provider | Endpoint | Notes | |----------|----------|-------| | DeepSeek | platform.deepseek.com | Best cost-effectiveness | | OpenAI | platform.openai.com | GPT-4o / GPT-4 | | Anthropic | console.anthropic.com | Claude 3.5 / 4 | | Any OpenAI-compatible | Custom base URL | Ollama, vLLM, etc. |

CLI Coding Backends

IdeaCo can dispatch coding tasks to local CLI assistants. Install any of these and they'll be auto-detected:

| Backend | Command | Description | |---------|---------|-------------| | Claude Code | claude | Anthropic's AI coding assistant | | Codex | codex | OpenAI's coding agent | | CodeBuddy | codebuddy | Tencent's AI coding assistant |

No API key or CLI tool is strictly required — the system falls back to a built-in rule engine, but agents will only give mechanical responses.

📁 Project Structure

IdeaCo/
├── src/
│   ├── app/                  # Next.js App Router + API Routes
│   ├── components/           # React UI (Pixel Office, Group Chat, Dashboards...)
│   ├── core/                 # Core engine
│   │   ├── agent/            # Agent layer (LLM / CLI / Web)
│   │   ├── employee/         # Employee layer (memory, lifecycle, skills)
│   │   ├── organization/     # Org layer (company, dept, team, HR)
│   │   ├── system/           # System services (audit, cron, plugins)
│   │   ├── prompts.js        # All LLM prompt templates
│   │   └── requirement.js    # Requirement workflow engine
│   ├── lib/                  # Frontend utilities (Zustand store, i18n)
│   └── locales/              # i18n (zh/en/ja/ko/es/de/fr)
├── data/                     # Runtime data (auto-created)
├── workspace/                # Agent-produced files per department
├── electron/                 # Electron desktop app shell
└── package.json