English | 简体中文 | 繁體中文 | 日本語 | 한국어 | Español | Türkçe | Русский | Tiếng Việt
A headless CLI system for installing, managing, and orchestrating AI agents, skills, and workflows. Designed so that AI agents can understand, install, and operate the entire system autonomously.
The system has evolved into a full-fledged 12-subsystem architecture:
- 💬 Chat — A unified conversation interface for the whole system. Multi-session threads with durable history, an agent picker (or automatic routing to the right specialist), and markdown rendering. Every turn runs the full pipeline — zero-token intent classification, skill selection, then the model — so anything the Telegram bot can do, the browser can do too.
- 📋 Codex — Mission control for agentic work. You or the AI create a task, delegate it to an agent or a team, approve it, and a background worker runs it while you watch the step timeline. Results come back for review. Tasks are durable and survive a restart, with a full audit trail per task.
- 🤖 Agent Manager — Create and manage AI agents with personas, routines, and skills.
- ⚡ Skill System — Executable workflows marked with tags (Workflow, API, Markdown) featuring a Markdown Viewer and Real-time Execution Modal.
- 🔄 Workflow Engine & Builder — DAG-based workflow executor. The WebUI features a modern node-based builder with compact nodes, contextual sliding property panels, and dynamic model selection (Ollama local / Cloud API).
- 🎨 Web Dashboard — Comprehensive SPA (Single Page Application) at
localhost:5295/dashboardto visually manage agents, workflows, skills, marketplace, settings, and monitor browsers natively. - 👥 Teams Agents — Orchestrate multiple agents using Organizational Charts. Assign roles via logical templates or drag-and-drop. Task Delegation routes work through the team based on sequential, parallel, or hierarchical strategies.
- 🏢 3D Studio (Teams 3D) — Isometric procedural 3D visualization using Three.js. Supports multi-seat furniture (meeting tables, conference tables) with intelligent inward-facing algorithms, raycasting group manipulation, and 15+ built-in assets.
- 🎬 Story Engine & Player — Generate interactive 3D stories from prompts via our Script Editor. Agents communicate via 3D speech bubbles inside an animated scene player.
- 🔌 Extension Manager — Pluggable architecture. Built-ins include
chat,codex,browser,browser_scripts,webui,market,multi_agents,studio3d,cloud_api, and more. Each extension contributes CLI commands, API routes, workflow nodes, Telegram actions, and its own UI page. - 🌐 Browser Automation — Orchestrate browser profiles, proxies, fingerprints. Built-in Auto-Login for Google with TOTP 2FA.
- 🛒 Marketplace — Discover, install, and share community skills via an online registry.
- 📨 Telegram Bridge — Drive the same system from a Telegram bot: intent routing, skill execution, task approval, and proactive notifications when long-running work finishes.
For Windows: Open PowerShell (Run as Administrator) and paste the following command:
powershell -c "irm https://raw.githubusercontent.com/tubecreate/tubecli/main/install.ps1 | iex"Note: To install with a specific language (e.g. vi, zh-TW, ja), use:
powershell -c "&([ScriptBlock]::Create((irm https://raw.githubusercontent.com/tubecreate/tubecli/main/install.ps1))) -Lang vi"For Linux / MacOS: Open your terminal and run:
curl -fsSL https://raw.githubusercontent.com/tubecreate/tubecli/main/install.sh | bashNote: To install with a specific language (e.g. vi, zh-TW, ja), use:
curl -fsSL https://raw.githubusercontent.com/tubecreate/tubecli/main/install.sh | bash -s -- --lang viIt will automatically install Python, Git (if missing), clone the repo, and set up everything for you from A-Z.
- Python 3.9+
- Ollama (Optional, required for local AI execution)
- Git
git clone https://github.com/tubecreate/tubecli.git
cd tubecli
pip install -e .Run the initialization command to setup the data/ directory, extract default skills, activate core extensions, and configure the default port.
tubecli init --lang en --port 5295After initialization, start the API server to access the GUI.
tubecli api startOpen your browser and navigate to: http://localhost:5295/dashboard
Manage the entire system directly from the terminal if you prefer a headless approach:
tubecli agent create "My Assistant" --description "General purpose AI agent"
tubecli agent list
tubecli agent show <id>
tubecli agent delete <id>tubecli skill list
tubecli skill run "AI Summarizer" --input "Long text content..."tubecli api start --port 5295
tubecli api stop
tubecli workflow run <path_to_workflow.json>tubecli codex create "Research the top 5 competitor channels" --agent "Researcher"
tubecli codex list --status pending_approval
tubecli codex approve 3
tubecli codex show 3Tasks the AI creates always wait for your approval before they run.
tubecli extension list
tubecli extension enable webui
tubecli market search "seo"
tubecli market install "seo-analyzer"Extensions bind their API routes when the server imports them, so restart the server after enabling or adding one.
tubecli/
├── tubecli/ # Main package
│ ├── api/ # REST API server (FastAPI)
│ ├── cli/ # CLI command modules
│ ├── core/ # Core Business logic
│ ├── extensions/ # Extensions (Chat, Codex, Browser, WebUI, Market, Studio3D, …)
│ ├── nodes/ # Workflow Node implementations
│ └── skills/ # Built-in system skills
├── .agents/ # AI-readable documentation (SKILL.md)
├── data/ # Runtime DB & State (gitignored)
└── tests/ # Test suite
The .agents/ and skills folders contain documentation crafted explicitly for LLMs (SKILL.md). External AI agents (like Claude or GPT-4) can read these files to learn how to operate the TubeCLI system, write plugins, and debug workflows completely autonomously without human intervention.
MIT License - Made with 🤖 by TubeCreate Team