Tubecli AI Agent

TubeCLI — Self-Hosted AI Agent Team That Makes Videos, Drives Browsers, Deploys Sites and Works a Task Board

Coding agents write code. This one runs the operations — on your machine, with your keys, for your clients.

<p align="center"> <b>English</b> | <a href="READMEs/README_zh-CN.md">简体中文</a> | <a href="READMEs/README_zh-TW.md">繁體中文</a> | <a href="READMEs/README_ja.md">日本語</a> | <a href="READMEs/README_ko.md">한국어</a> | <a href="READMEs/README_es.md">Español</a> | <a href="READMEs/README_tr.md">Türkçe</a> | <a href="READMEs/README_ru.md">Русский</a> | <a href="READMEs/README_vi.md">Tiếng Việt</a> </p> <p align="center"> <a href="https://github.com/tubecreate/tubecli/stargazers"><img src="https://img.shields.io/github/stars/tubecreate/tubecli?style=for-the-badge&color=2a2a2a&labelColor=1a1a1a" alt="GitHub stars for TubeCLI" /></a> <a href="https://github.com/tubecreate/tubecli/blob/main/LICENSE"><img src="https://img.shields.io/badge/LICENSE-MIT-00897b?style=for-the-badge&labelColor=333333" alt="MIT License — free to resell and white-label" /></a> <img src="https://img.shields.io/badge/SELF--HOSTED-YES-4c1?style=for-the-badge&labelColor=333333" alt="Self-hosted AI agents, no cloud required" /> <img src="https://img.shields.io/badge/LOCAL_LLM-OLLAMA-ff6b35?style=for-the-badge&labelColor=333333" alt="Runs on local LLM via Ollama" /> <img src="https://img.shields.io/badge/PYTHON-3.10+-0078d4?style=for-the-badge&logo=python&logoColor=white&labelColor=333333" alt="Python 3.10+" /> <img src="https://img.shields.io/badge/API-FASTAPI-009688?style=for-the-badge&logo=fastapi&logoColor=white&labelColor=333333" alt="FastAPI backend" /> </p>

TubeCLI is a self-hosted, open-source AI agent team that executes real operational work end to end. It is not a chat wrapper, and it is not a framework you have to code your agents into. You describe the job in plain language — from the CLI, the web dashboard, Telegram, or the Flow canvas on TubeCLI Cloud — and an agent team carries it out with real tools against your own machine, with your own keys.

v2026.08.09.193 · Python 3.10+ / FastAPI / Node.js / Vue / Three.js · MIT

---

What's new (August – October 2026)

• Narrated videos from what the agent read today. content_video turns the agent's own corpus (articles it scraped, videos it watched, links you paste) into a script → storyboard → images → voice → mp4, queued on the Task Board. Aspect 16:9 or 9:16, any of nine languages, verbatim mode for scripts you wrote yourself, a content queue that feeds one video per item per day, resume-after-crash rendering, and optional auto-publish to YouTube with a generated thumbnail.
• Canvas Engine in the core. The scene renderer (layouts, word-timed subtitles, chalkboard and sketch templates, per-scene clips) now ships with TubeCLI and updates with it.
• Codex GPT node. The OpenAI Codex CLI runs on your machine, under TubeCLI: several ChatGPT subscriptions in one vault with automatic switching when a quota runs out, chat sessions managed from the Flow canvas, and an MCP server (tubecli) through which Codex drives the Task Board, browser profiles, extensions and the API.
• Muse as a provider. muse.ai (Meta's agent) has no API; TubeCLI drives its web app inside a signed-in browser profile and exposes it like any other provider — text, 16:9 / 9:16 images, 10-second image-to-video clips, several accounts in parallel.
• More voices and images. OmniVoice and EverAI voices, TTS VibeVoice (Edge fallback), CapCut TTS with word timings; one shared image core (core/image_gen.py) with Cloudflare FLUX, gpt-image, Gemini and Muse behind it, daily-quota key rotation and a "test this key" button.
• Deliverables where you keep them. Finished videos, per-scene clips, SRT subtitles, director boards and a Google Sheet of prompts go to your Google Drive; clone a finished video into another language with one click; retry a step with a different writing model, image model or voice.
• TubeCLI Cloud and Agent Town. Optional hosted canvas at cloud.tubecreate.com: pair your own PC with a 6-character code (free), get a VPS from the platform, or connect one you already have over SSH; watch browsers live; and publish your agents on Agent Town where visitors chat with them and hire them — ad videos from product photos, Vietnamese "NĐ 30" Word formatting, remote browser sessions.
• Your own PC as the server. A one-file connect client installs TubeCLI if missing, keeps it running, holds a Cloudflare tunnel, sits in the tray and starts with the computer. Windows, macOS and Linux — including machines in mainland China via install-cn.sh and a GitHub proxy.
• Operations. Extensions update by hot reload without restarting TubeCLI; machines report CPU / RAM / disk to the cloud every 15 minutes; browser sessions recover from "already open" and out-of-memory states with a named reason instead of a silent close.

---

What it actually does

Make a video from today's reading

Tell the agent "make a video from what you read today" (or hand it links), and it files a task:

{"action": "content_video_run", "aspect_ratio": "9:16", "language": "en", "target_words": 1500,
 "sources": ["https://example.com/article", "https://youtube.com/watch?v=..."],
 "publish": true, "publish_channel_name": "My Channel", "thumbnail": true}

What runs, in order, with every step on the Task Board and a resumable render: check that Content Studio has a text and an image provider (fail here, before spending anything) → read the agent's corpus for the day → fetch transcripts and crawl the extra links → write the script with the agent's model → storyboard with a Content Studio template → draw the images → read the narration (CapCut, VibeVoice, OmniVoice, EverAI or Edge voices) → render with the Canvas Engine → upload to Drive, and publish to YouTube if you said so. Paste your own text and the video reads it verbatim, the AI only describes the pictures.

Deploy a real website from one sentence

Say it in chat or Telegram — "create a website called coffee-shop using the coffee-machine template" — and the website agent runs the entire chain against your own Cloudflare account, streaming every step to a live log:

[1/9] git clone --depth=1 -- <template>            → clone
[2/9] npm install
[3/9] wrangler d1 create + wrangler r2 bucket create
[4/9] wrangler d1 execute schema.sql --remote       → real remote database
[5/9] write wrangler.toml  (D1 + R2 bindings)
[6/9] npx @opennextjs/cloudflare build
[7/9] npx wrangler deploy                          → https://<site>.workers.dev
[8/9] seed the admin password you chose
[9/9] DEPLOY_RESULT {url, adminSeeded}

Every step is in deploy_runner.js — read it before you trust it. It is not one-click: it needs valid Cloudflare credentials and D1 quota, and when a step fails it stops and names the step in the log rather than reporting a half-built site as finished.

Subtitle, translate, dub and clone a video

tubecli skill run "Analyze Channel" --input "https://www.youtube.com/@channel"
tubecli skill run "Extract Subtitle" --input "<video-url-or-path>"
tubecli skill run "Translate Subtitle" --input "<job-id>"

Three interchangeable extraction engines — local Whisper, Gemini, or YouTube captions — exporting SRT / JSON / VTT / ASS. A finished TubeCLI video can be cloned into another language (new narration, same pictures) from its Task Board card. Burn-in and hardsub removal (ffmpeg delogo/blur/pixel/fill with OpenCV region detection) are implemented and reviewable, but treat them as experimental.

Drive a real browser

Anti-detect browser profiles (ShardX engine on Linux/macOS, BAS on Windows), proxies and fingerprints per profile, a Keychain vault so the agent logs in without ever seeing your password, TOTP 2FA, YouTube cookies fetched from a running profile when a download needs them, and Script Studio for recorded step-by-step automations. On the Flow canvas you watch the browser live, type into it, and hand it to an agent — or let an agent run it headless on a schedule and peek in when you want.

Keep a human in front of anything irreversible

An agent proposing something destructive does not just do it:

codex                  # list what is waiting
codex 3                # inspect task 3 — goal, assignee, step timeline
approve 3              # or:  reject 3

Tasks the AI creates land in pending_approval and wait. Deploying, publishing and deleting all pass through that gate, and every task keeps an append-only JSONL audit trail with step timings and retry counters.

Run it from wherever you are

The same pipeline backs every surface — CLI, web dashboard, Telegram bot, and the Flow canvas on TubeCLI Cloud. Send a link to the bot and it replies with the finished file; approve a task from your phone; watch a render or a deploy log stream in the browser.

Models: every LLM call runs on local Ollama — no API key, no per-token cost — or on Gemini, OpenAI, Claude, DeepSeek, Grok, OpenRouter, Cloudflare Workers AI, a local 9Router endpoint, Muse (through a browser session) and the Codex CLI (your ChatGPT plan) with your own keys. Critical paths use zero-token fast paths that never call a model at all.

Also included: WordPress publishing over the WP REST API, web crawling and change watching, video download from 50+ platforms, Douyin/TikTok download, Google Sheets, Calendar and Drive, a shared media library, POD / graphic / content studios, and livestream restreaming.

---

The dashboard

All of it is also driveable from a browser at localhost:5295/dashboard — agents, browser profiles, skills, workflows, extensions, API keys, the Task Board and the Codex GPT console in one place, with local Ollama and browser status live in the header and every installed studio in the sidebar.

---

Why people pick it

• "It generated a scaffold, not a website." The deploy runner executes nine real steps against your Cloudflare account and does not stop until a URL answers. You can read every step in the repo. See the caveat above — it is real, not magic.
• "I want the video, not a script for a video." content_video ends with an mp4 on your Drive (and on your channel if you asked). Steps that fail say so on the card; a killed render resumes from the frames it already has.
• "The framework is free, but I'm not allowed to resell it." TubeCLI is MIT: multi-tenant deployment, white-labelling and reselling are all permitted. For contrast, n8n's Sustainable Use License forbids reselling it as a hosted service, and Dify's modified Apache license forbids operating a multi-tenant environment and forbids removing their logo. (CrewAI, LangGraph and OpenHands are MIT too — this argument does not apply to them.)
• "My agent burned tokens deciding what to do." Common flows are routed by zero-token fast paths — deterministic intent matching that never calls a model. The LLM is used where judgement is actually needed.
• "I don't want my clients' credentials in someone else's cloud." Everything runs on your hardware. Your keys stay in your own data/ directory, social logins stay in the Keychain vault, and with local Ollama the system needs no external API at all.
• "An agent did something irreversible." The Codex queue gates it: AI-created tasks land in pending_approval and wait for a human before anything irreversible runs — deploy, publish, delete. Every task carries an append-only JSONL audit trail, and a running job can be cancelled mid-flight and restarted from the step it stopped at.

🌟 Key Features

• 💬 Chat — One conversation surface for the whole system: multi-session threads with durable history, an agent picker (or automatic routing), 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.
• 📋 Task Board (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 with progress, elapsed time and time left. Extensions register their own task types (a video, a queue item, a reup) with their own forms. Tasks survive a restart and carry a full audit trail.
• 🎬 Content video pipeline — Script → storyboard → images → voice → render → Drive → YouTube, from the agent's corpus or from text you paste. Content Studio templates decide language, voice, layout and length; a shared template library serves Content Studio and POD Studio alike.
• 🖼 Canvas Engine — The renderer behind "explainer" videos: scene layouts, word-timed subtitles (.ass), chalkboard / sketch / watercolor kits, per-scene clean clips, 16:9 and 9:16.
• 🧠 Codex GPT — Codex CLI managed by TubeCLI: multiple ChatGPT subscriptions with automatic failover, sessions in the Flow canvas, and the tubecli MCP server so Codex can operate TubeCLI itself.
• 🎨 Muse provider — Text, images and 10-second clips from muse.ai through a signed-in browser profile; several accounts run in parallel lanes.
• 🤖 Agent Manager — Agents with personas, routines, skills, per-agent thresholds (busy / tired / quota warnings) and a public mode for Agent Town.
• ⚡ Skill System — Executable workflows marked with tags (Workflow, API, Markdown) with a Markdown viewer and a real-time execution modal.
• 🔄 Workflow Engine & Builder — DAG-based executor with a node-based builder in the WebUI; the Flow canvas on TubeCLI Cloud adds live browser, agent, extension, Codex and file nodes.
• 👥 Teams Agents — Orchestrate agents with organizational charts; delegation routes work sequentially, in parallel, or hierarchically.
• 🏢 3D Studio & Story Engine — Isometric procedural 3D teams and interactive 3D stories (Three.js).
• 🌐 Browser Automation — Profiles, proxies, fingerprints, Keychain logins, TOTP 2FA, Script Studio, live view, headless scheduled runs, orphan and out-of-memory recovery.
• 🔐 Keychain — Your social-media logins in a Fernet-encrypted vault; browser profiles use them, agents never read them.
• 📚 Media Library — One bag of images, GIFs and videos every extension can draw from.
• 🔌 Extension Manager — Pluggable architecture; each extension contributes CLI commands, API routes, workflow nodes, Telegram actions, task types and its own UI page. Updates hot-reload.
• 🛒 Marketplace — Install studios and skills from the registry in one click, and publish your own templates.
• 📨 Telegram Bridge — Intent routing, skill execution, task approval and notifications from a bot.
• ☁️ TubeCLI Cloud (optional) — Hosted Flow canvas, pairing for your own PC, VPS rental, SSH connect, metrics, and Agent Town.

🧩 What ships in the repo

26 built-in extensions (tubecli/extensions/):

Extension	What it does
chat	Conversation surface and the intent → skill → model pipeline
codex	Task Board: durable queue, approval gate, background worker, audit trail
codex_gpt	Codex CLI under TubeCLI: subscriptions vault, sessions, tubecli MCP server
content_video	Narrated video from the agent's corpus or pasted text, as a task
content_queue	Keyword + text items, one video each, with a daily cap
canvas_engine	Scene renderer: layouts, subtitles, templates, per-scene clips
capcut_tts	CapCut voices with word timings, account pool
media_library	Shared raw material (images, GIFs, videos)
browser	Anti-detect profiles, proxies, fingerprints, live view, Muse driver
browser_scripts	Script Studio: visual step-by-step browser automations
keychain	Encrypted vault for the user's social logins
auth_manager	OAuth credentials and tokens for Google, Facebook, TikTok
website_manager	Cloudflare Workers deploy runner, site records, WordPress publishing
video_studio	Subtitle extraction / translation, channel analysis, approval-gated reup
video_downloader	yt-dlp with self-install and self-update, cookies from a running profile
video_editor	FFmpeg trimming, merging, overlays, export
douyin_downloader	Douyin / TikTok download with direct video links
universal_tracker	Watches YouTube, Douyin and websites for new posts and runs team workflows
multi_agents	Teams, org charts, delegation strategies
cloud_api	Provider keys: Gemini, OpenAI, Claude, DeepSeek, Grok, OpenRouter, Cloudflare, 9Router, Muse
ollama_manager	Local models with no API key
market	Marketplace client: install, update, publish
calendar_manager	Google Calendar events, recurrence, reminders
file_manager	Files and folders on the machine, Google Drive, share links
studio3d	3D teams and story player
webui	The dashboard itself

Marketplace studios (install from the dashboard): Content Studio, EduVideo Studio, POD Ad Studio, Thumbnail Studio, Web Crawler, Subtitle Extractor, TTS VibeVoice, Sheets Manager, Video Manager, Browser Scripts, AI Arena, livestream, and more. Content Studio and TTS VibeVoice come with every TubeCLI: on first start the machine installs them from the Market in the background, once. If you uninstall one, it stays uninstalled. Set TUBECLI_NO_DEFAULT_EXTENSIONS=1 to skip this.

☁️ TubeCLI Cloud and Agent Town

TubeCLI runs fine on its own. TubeCLI Cloud is an optional hosted layer in front of it:

• Flow canvas — a shared workspace of nodes: browser profiles you watch live, agents, extensions, the Task Board, Codex GPT sessions, files and sheets. A guided start for newcomers picks the AI model, installs the studios, checks ffmpeg and voices, and creates the first task.
• Three ways to add a machine — Connect my computer (free: a pairing code and one line to paste; the open-source client in client/tubecli_connect.pyw installs TubeCLI, starts it and holds a Cloudflare tunnel), Get a server (we host it, monthly), or Connect my VPS (we install over SSH on a Linux box you own).
• Agent Town — publish an agent and visitors chat with it, hire it for ad videos from their product photos, Word formatting to Vietnamese decree NĐ 30, or a remote browser session, and pay in credits. The machine reports progress with HMAC-signed telemetry; the owner always approves what runs.

The cloud never holds your keys or your browser sessions; it reaches your machine only through the tunnel you opened.

🚀 Quick Start & Installation

Option 1: One-Click Auto Install (Recommended for Users)

For Windows: Open PowerShell and paste the following command:

powershell -c "irm https://raw.githubusercontent.com/tubecreate/tubecli/main/install.ps1 | iex"

To install with a specific language (e.g. vi, zh-TW, ja):

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 | bash

With a specific language:

curl -fsSL https://raw.githubusercontent.com/tubecreate/tubecli/main/install.sh | bash -s -- --lang vi

Mainland China (Linux server): raw.githubusercontent.com is reset there, so use the China installer through a GitHub proxy — it also switches Node, npm and pip to local mirrors and fetches cloudflared through the proxy:

curl -fsSL https://gh-proxy.com/https://raw.githubusercontent.com/tubecreate/tubecli/main/install-cn.sh | bash

On Windows and macOS in China, pass the proxy to the normal installers: install.ps1 -RepoUrl https://gh-proxy.com/https://github.com/tubecreate/tubecli.git, or TUBECLI_REPO_URL=https://gh-proxy.com/https://github.com/tubecreate/tubecli.git for install.sh. TubeCLI Cloud prints the right line for you when you switch on Mainland China network.

The installer adds Python, Git and Node.js if missing, clones the repo, and sets everything up from A to Z.

On a server (VPS): the same command detects that there is no display and finishes as a server — it asks for a dashboard password (Enter keeps the default 123456; until you change it — at /login or with tubecli password — a remote login can do nothing but change it), installs a systemd service so TubeCLI survives SSH logout and reboots, and prints the URL to open from your own computer. Re-running the same command later updates the install; the dashboard also has a one-click Update button. Useful afterwards:

tubecli info        # re-print the URL / password status / commands screen
tubecli password    # change the dashboard password
tubecli service logs -f

Option 2: Manual Installation (For Developers)

Prerequisites

• Python 3.10+
• Git
• Node.js 18+ and ffmpeg (the installers add them; the browser engine, Canvas Engine and Codex GPT need Node)
• Ollama (optional, for local models)

1. Clone & Install

git clone https://github.com/tubecreate/tubecli.git
cd tubecli
python3 -m venv .venv && . .venv/bin/activate   # macOS / Linux
pip install -e .

2. Initialize Workspace

init sets up the data/ directory, extracts default skills, activates core extensions and configures the port. On first run it also walks you through picking an AI model.

tubecli init --lang en --port 5295

It then hands over to an interactive control panel, which starts the API server for you and stays in the foreground — press 1 to open the dashboard. This is the normal way to run TubeCLI day to day.

3. Or run it as a server

On a headless machine (VPS, no display) tubecli init skips the control panel automatically and finishes as a server: password, systemd service, and a summary with the URL. You can also force either ending:

tubecli init --server   # password + systemd service + summary, then exit
tubecli init --panel    # the interactive control panel, even on a server
tubecli init --no-menu  # plain setup and exit (Dockerfile / CI; then run `tubecli api start`)

On a desktop the dashboard is at http://localhost:5295/dashboard; on a server replace localhost with the machine's public IP (and open TCP 5295 in your provider's firewall). tubecli info prints the exact URL block any time.

Option 3: Pair your own computer with TubeCLI Cloud

Open cloud.tubecreate.com → Connect my computer, pick your system and get a pairing code. The page gives you one line to paste; this is what it does on Windows:

powershell -c "irm https://raw.githubusercontent.com/tubecreate/tubecli/main/client/tubecli_connect.pyw -OutFile $env:TEMP\tubecli_connect.pyw; pythonw $env:TEMP\tubecli_connect.pyw --code=XXXXXX"

The client (one file, standard library only) installs TubeCLI if it is missing, starts it on port 5295, opens a Cloudflare tunnel to <name>.tubecreate.com, sits in the tray and starts with the computer. Already installed by hand? Pair with the client that is in the repo: python client/tubecli_connect.pyw --code=XXXXXX. To remove it: tray icon → Quit, then delete ~/TubeCLI (Windows) or ~/tubecli and the config folder. Nothing is charged for your own machine.

💻 CLI Usage

Manage the entire system directly from the terminal if you prefer a headless approach:

Agent Management

tubecli agent create "My Assistant" --description "General purpose AI agent"
tubecli agent list
tubecli agent show <id>
tubecli agent delete <id>

Skill Execution

tubecli skill list
tubecli skill run "AI Summarizer" --input "Long text content..."

API & Workflows

tubecli api start --port 5295
tubecli api stop
tubecli workflow run <path_to_workflow.json>

Task Board (Codex)

tubecli codex create "Research the top 5 competitor channels" --agent "Researcher"
tubecli codex list --status pending_approval
tubecli codex approve 3
tubecli codex show 3

Tasks the AI creates always wait for your approval before they run.

Extensions & Market

tubecli extension list
tubecli extension enable webui
tubecli market search "seo"
tubecli market install "seo-analyzer"

Extensions installed or updated from the dashboard hot-reload. Extensions you drop into the folder by hand bind their routes when the server imports them, so restart the server after adding one that way.

How TubeCLI compares

TubeCLI column measured 2026-10-09; competitor columns were last checked 2026-07-30 and may have moved. Rows where TubeCLI loses are marked plainly — if you need those, use one of the others.

	TubeCLI	n8n	Dify	CrewAI	LangGraph	OpenHands
What it sells	Work already finished	Workflow platform	Agentic workflow builder	Agent framework	Low-level orchestration	Coding-agent control center
License	MIT	Sustainable Use	Modified Apache	MIT	MIT	MIT
Resell as a hosted service	Yes	No	No	Yes	Yes	Yes
Multi-tenant / white-label	Yes	No	No (logo must stay)	Yes	Yes	Yes
Self-host	Yes	Yes	Yes	Yes	Yes	Yes
Local LLM first-class (Ollama)	Yes	Partial	Partial	Via config	Via config	Partial
Zero-token fast paths	Yes	n/a	No	No	No	No
Makes a finished narrated video (script → images → voice → mp4 → YouTube)	Yes	No	No	No	No	No
Deploys a site to Cloudflare Workers end-to-end	Yes (see caveat)	No	No	No	No	No
Crawl → draft with an LLM → publish to WordPress	Yes	Build it yourself	Build it yourself	Build it yourself	Build it yourself	No
Subtitle extract + translate (3 engines, SRT/VTT/ASS)	Yes	No	No	No	No	No
Browser automation with live view, Keychain logins, TOTP 2FA	Yes (2FA flaky)	Via nodes	No	No	No	No
Human approval gate on irreversible steps	Yes, enforced	Manual	No	No	Build it yourself	Partial
Telegram as a control plane	Yes	Via nodes	No	No	No	No
No-code visual builder	Partial — workflow builder in the dashboard, Flow canvas on TubeCLI Cloud	Yes	Yes	No	No	Partial
MCP support	Partial — ships an MCP server (tubecli) for the Codex CLI; no generic MCP client	Yes	Yes	Yes	Yes	Yes
Prebuilt integrations	26 built-in + marketplace studios	400+	Large	Large	Large	Large
Managed cloud option	Optional (TubeCLI Cloud)	Yes	Yes	Yes	Yes	Yes
Enterprise support / SSO / audit	No	Yes	Yes	Yes	Yes	Partial
Battle-tested at scale	No — early	Yes	Yes	Yes	Yes	Yes

Caveat on Cloudflare deploys: real but not one-click — it needs valid Cloudflare credentials and D1 quota, and admin-password seeding can fall back to the template default. Failures stop and name the failing step instead of reporting success.

Where TubeCLI honestly loses: no generic MCP client, a thinner no-code builder than n8n or Dify, no enterprise SSO/audit product, and far fewer integrations than n8n's 400+. It is early software.

Where the license comparison does not apply: CrewAI, LangGraph and OpenHands are MIT too — the resale argument only beats n8n, Dify and AutoGPT's platform. Against CrewAI and LangGraph the difference is that they hand you building blocks, while TubeCLI hands you a wrangler deploy that already ran and an mp4 that is already on your Drive.

FAQ

What is TubeCLI?

TubeCLI is a self-hosted, open-source AI agent team that executes operational work end to end — making narrated videos from what the agent read, deploying websites to Cloudflare Workers, crawling the web and publishing to WordPress, extracting and translating video subtitles, and automating browsers. It ships 26 built-in extensions plus marketplace studios, and runs from the CLI, a web dashboard, Telegram, or the Flow canvas on TubeCLI Cloud.

Is TubeCLI free?

Yes. MIT licensed, no paid tier, no seat limits, no credit meter. You may run it for paying clients, multi-tenant and white-labelled, without asking permission. Your only costs are your own infrastructure and whatever LLM you point it at — which can be $0 if you use local Ollama. TubeCLI Cloud is a separate, optional hosted service; pairing your own computer with it is free.

Does TubeCLI need an API key?

No. Point it at local Ollama and it runs with no API key and no per-token cost. If you prefer cloud models it supports Gemini, OpenAI, Claude, DeepSeek, Grok, OpenRouter and Cloudflare Workers AI with your own keys, a local 9Router endpoint, Muse through a signed-in browser profile, and the Codex CLI with your ChatGPT plan. Note that the shipped default is a cloud model, so switch the default to Ollama if you want fully local operation.

Can it really make a video by itself?

Yes, as a task you can watch and stop. content_video writes the script from the agent's corpus (or reads your text verbatim), storyboards it with a Content Studio template, draws the images, reads the narration, renders with the Canvas Engine and uploads the result — to YouTube too if you asked, with a generated thumbnail. It needs Content Studio (installed on first start) with a text and an image provider, ffmpeg, and a voice; content_video_capabilities lists exactly what is missing before anything is spent. A render that was killed resumes from the frames it already has.

What is the Codex GPT node, and how is it different from the Task Board?

The Task Board (codex) is TubeCLI's own queue of tasks for its agents. Codex GPT (codex_gpt) is OpenAI's Codex CLI running on your machine under TubeCLI's management: a vault of several ChatGPT subscriptions that switch automatically when one runs out of quota, chat sessions you manage from the Flow canvas, and an MCP server named tubecli through which Codex can read the Task Board, open browser profiles, install extensions and call the API — with changes still waiting for the owner's approval.

Can TubeCLI actually deploy a website?

Yes, genuinely — not a scaffold. Its deploy runner performs git clone → npm install → wrangler d1 create + r2 bucket create → remote schema apply → OpenNext build → wrangler deploy against your own Cloudflare account, and you can read the whole runner in this repo before trusting it. It is not one-click: it needs valid Cloudflare credentials and D1 quota, and admin seeding can leave the template default password. When something fails it stops and names the failing step in the log instead of reporting a half-built site as finished.

How is TubeCLI different from LangChain, CrewAI, or AutoGPT?

Those give you building blocks to construct an agent; you still write the supervisor, the deploy step, and the publishing step yourself. TubeCLI ships the finished tools — a video pipeline that ends with an mp4 on your Drive, a Cloudflare deploy runner, a WordPress publisher that speaks WP REST with app-password auth, a three-engine subtitle pipeline. You operate it rather than program it.

Is TubeCLI a self-hosted Manus alternative?

For operational tasks, yes — both aim at an agent that uses real tools rather than returning code. The differences that matter: TubeCLI runs on your hardware, has no credit meter, is MIT so you can resell it, and can run entirely on a local model. It is far less polished and far less battle-tested than a funded hosted product.

Can I run TubeCLI for my clients and charge for it?

Yes, and this is a deliberate differentiator. MIT permits multi-tenant deployment, white-labelling, and reselling. For contrast: n8n's Sustainable Use License forbids reselling it as a hosted service, and Dify's modified Apache license forbids operating a multi-tenant environment and forbids removing their logo. Agent Town goes one step further: publish your agent there and visitors hire it directly.

Does TubeCLI support MCP?

Partially. The repo ships an MCP server (tubecli/extensions/codex_gpt/mcp_relay.py, registered as tubecli in the Codex CLI's config.toml) that exposes the Task Board, browser profiles, extensions and the API to Codex. There is no generic MCP client, so TubeCLI cannot yet consume third-party MCP servers; if that is a requirement, use n8n, Dify, CrewAI or LangGraph instead.

Can Claude share files from my server?

Yes, when you allow it. Connect claude.ai to TubeCLI Cloud (Connectors), then tick Allow public share links for that connection. Ticking the box is your consent in advance: Claude's share_file / share_files make a link at once, with no per-link approval. Each link gives a page, a raw URL and a direct download URL that work with a plain HTTP GET (no sign-in; Range supported) and stop after 30 minutes unless Claude asks for a different expiry (1–1440 minutes).

• Only images (.png .jpg .jpeg .webp .gif) and videos (.mp4 .webm .mov) inside the TubeCLI data folder can be shared. The file's content is checked as well as its extension, and checked again every time the link is opened.
• Browser profiles, sessions, cookies, keys and .env files are refused, and so are paths that leave the data folder (.., symlinks).
• A connection can make at most 200 links an hour. Every link created or downloaded is logged in data/file_manager/share_log.jsonl (time, connection, file, IP).
• Expired or revoked links answer 410 Gone. Unticking the box, disconnecting Claude or turning Claude off on the server revokes that connection's live links at once. You can also revoke any link in File Manager › Shares.

Can TubeCLI translate and burn subtitles?

Extraction and translation are the mature part: three interchangeable engines (local Whisper, Gemini, YouTube CC) exporting SRT/JSON/VTT/ASS, with translation as a separate step so you can review the text before it is burned anywhere. Finished TubeCLI videos can be cloned into another language. Burn-in and hardsub removal (ffmpeg delogo/blur/pixel/fill with OpenCV region detection) are implemented and reviewable but should be treated as experimental.

How do I stop an agent from doing something irreversible?

The Codex queue gates it. AI-created tasks land in pending_approval and wait for a human to accept or reject before anything irreversible runs — deploying, publishing, deleting. Review it with codex 3, then approve 3 or reject 3. Every task carries an append-only JSONL audit trail with step timings and retry counters, and a running job can be cancelled mid-flight.

Is TubeCLI production-ready?

Parts of it are. Browser automation, subtitle extraction and translation, WordPress publishing, the content video pipeline and the Task Board are in daily use. Cloudflare deploys work but need supervision. The full video reup chain, subtitle burn-in, Muse video clips and multi-agent org-chart delegation are implemented but should be treated as experimental. Treat it as capable early software with an honest changelog, not as enterprise infrastructure.

🧠 Architecture Overview

tubecli/
├── tubecli/           # Main package
│   ├── api/           # REST API server (FastAPI) + Codex GPT and Muse routes
│   ├── cli/           # CLI command modules
│   ├── core/          # Core business logic: brain, image_gen, muse, telegram, net_mirror…
│   ├── extensions/    # 26 built-in extensions (each with its own SKILL.md or README)
│   ├── nodes/         # Workflow node implementations
│   └── skills/        # Built-in system skills
├── client/            # tubecli_connect.pyw — pairing client for TubeCLI Cloud (one file)
├── install.sh / install.ps1 / install-cn.sh
├── READMEs/           # Translated documentation (8 languages)
├── docs/              # Project website
├── data/              # Runtime DB & state (gitignored)
└── tests/             # Test suite

📖 AI-Readable Documentation

TubeCLI is designed so that an AI agent can understand and operate it without a human walkthrough:

• llms.txt at the repository root — a compact, machine-readable map of what the system is, what it can do, and what it explicitly cannot do.
• SKILL.md inside every extension (tubecli/extensions/*/SKILL.md) — endpoints, parameters, trigger commands and worked examples, written for LLMs rather than humans.
• SKILL_EXTENSION_BUILDER.md — how to write your own extension or skill.
• The tubecli MCP server — lets a Codex CLI session read the Task Board and operate the machine with the owner's approval.

External agents (Claude, GPT, Gemini, Codex) can read these files to learn how to drive the system, write extensions, and debug workflows autonomously.

🔒 Security

TubeCLI runs code and holds your logins, so it is built around one rule: the machine owner is fully trusted, and nobody else may become the owner. SECURITY.md lists every layer that enforces it (login gate and origin guard that fail closed, proxy-aware loopback rule, scoped guest tokens, signed public-agent calls, the approval gate), what is stored where and with which permissions, and the residual risks we state rather than hide — the default-password window, the 0.0.0.0 bind on headless servers, and extensions running in-process without a sandbox among them. Pin an extension to the commit you reviewed with tubecli extension install <url> --commit <full hash>; every install is recorded in data/extension_provenance.json. SECURITY.md also shows how to download and read the install scripts before running them, and how to report a vulnerability privately.

📝 License

MIT — you may use, modify, self-host, white-label and resell this software, including for paying clients. Made with 🤖 by the TubeCreate Team.