TubeAssistant — AI Agent for YouTube

Fully autonomous AI agent that runs your YouTube channel 24/7 — script, voice, editing, thumbnail, upload. Zero human input.

▶ Watch the full demo with sound

TubeAssistant is an open-source AI agent that automates your entire YouTube workflow. It generates scripts with an LLM, synthesizes narration, downloads stock footage, edits the video, creates an AI thumbnail, and uploads to YouTube — every day, automatically, at the time you choose.

No editing software. No recording. No scheduling. The agent runs 24/7 and you control it via Telegram.

The Problem

Growing a YouTube channel is a second job. Every single day you need to:

• Research what topics are working
• Write a full script
• Record or synthesize a voice-over
• Find and edit video footage
• Design a thumbnail
• Write SEO-optimized titles, descriptions, and tags
• Upload and schedule at the right time
• Analyze performance and adapt

Most creators quit within 3 months — not because they lack ideas, but because the process is relentless and eats hours every day.

TubeAssistant eliminates every step. It runs 24/7, reads your channel analytics, generates content, and publishes — completely autonomously.

Features

• Full pipeline automation — from topic idea to published video, zero human input
• AI script generation — LLM writes the script based on your niche and what's performing
• Natural voice narration — Edge TTS, no API cost, no robotic voice
• Stock footage — auto-downloads relevant clips from Pexels (free, millions of clips)
• AI video generation (optional) — generate the clips with Replicate, fal.ai, Luma or
  OpenRouter instead of downloading them, with a per-video spend cap. OpenRouter has no
  text-to-video model: it generates the scene as an image and the clip is built locally with a
  camera move — a fraction of the cost, and it reuses the key you already use for scripts.
  Set it from the TUI: tube-assistant clips
• Smart video editing — FFmpeg montage with synced audio and background music
• AI thumbnail — generated by Pollinations.ai or HuggingFace FLUX with automatic text overlay
• Analytics-driven — reads your channel data, adapts topic and tone accordingly
• Telegram control — send a message, the agent obeys. Stop, force, queue topics, check stats
• Crash recovery — pipeline checkpoints resume from the last completed step
• 100% free tier — OpenRouter, Pexels, Edge TTS, Pollinations.ai — all free

Quick Start

pip install uv                                                    # fast installer (one-time)
uv tool install git+https://github.com/metiu1/tube-assistant.git
tube-assistant             # interactive menu — pick Setup wizard first

That's it. The wizard handles API keys, Google OAuth, Telegram bot, and channel preferences.

How It Works

Analytics → Strategy → AI Script → TTS Voice → Stock Clips → Montage → AI Thumbnail → Upload

Every day, at the time you choose, TubeAssistant:

1. Reads your channel analytics to understand what's performing
2. Picks a topic aligned with your niche and goals
3. Writes a full script using an LLM (OpenRouter, Ollama Cloud, or local)
4. Synthesizes narration with Edge TTS — natural voice, zero cost
5. Gets the footage — stock clips from Pexels, or generated by an AI video model
   (Replicate / fal.ai / Luma) when you enable it
6. Edits the video with FFmpeg — synced audio, background music
7. Generates a thumbnail with AI (Pollinations.ai or HuggingFace FLUX) + automatic text overlay
8. Uploads to YouTube and schedules at the optimal time

Everything controllable in real time via Telegram — just message your bot.

Requirements

• Python 3.11+
• FFmpeg in PATH (download)
• Google account with YouTube Data API enabled
• Telegram bot (free, 2 min via @BotFather)
• OpenRouter API key (free tier works)
• Pexels API key (free)

Installation

# Clone
git clone https://github.com/metiu1/tube-assistant.git
cd tube-assistant

# Install
pip install uv              # one-time
uv tool install -e .        # or: installa.bat on Windows

# Setup (one-time wizard)
tube-assistant onboard

# Run
tube-assistant start

The onboard wizard walks through every step:

Step 1 — Agent language  Any language your Telegram assistant will speak
Step 2 — AI Service      OpenRouter / Ollama Cloud / local Ollama / 12 more
Step 3 — Image provider  Pollinations (free) or HuggingFace FLUX
Step 4 — Pexels API      Free key for stock footage
Step 5 — Telegram Bot    Token from @BotFather
Step 6 — Channel setup   Name, niche, goals, style, language
                          → Send your Google credentials.json

Telegram Control

Natural language works too:

"Next video about the future of remote work"
"Avoid mentioning crypto"
"Make the pacing faster"

AI Services

15 providers supported. Set AI_SERVICE in .env to switch at any time.

OpenRouter and Ollama (the wizard defaults) need no extra packages. Providers
with their own SDK are optional extras - install only what you use:

pip install "tube-assistant[openai]"     # also deepseek/xai/together/perplexity/fireworks/azure
pip install "tube-assistant[anthropic]"  # or: gemini, mistral, groq, cohere, huggingface, all

Every model is overridable via env (e.g. OPENAI_MODEL=gpt-4o, GROQ_MODEL=mixtral-8x7b-32768).
See .env.example for the full list of variables.

CLI

tube-assistant              # interactive TUI menu (recommended)
tube-assistant onboard      # first-time setup wizard
tube-assistant start        # start the 24/7 daemon
tube-assistant run          # one-shot pipeline (one video now)
tube-assistant clips        # clip source: Pexels stock or AI video generation
tube-assistant update       # update to latest version
tube-assistant status       # show workspace and config
tube-assistant workspace    # print workspace path

Cost

Total: $0/month on free tiers.

If you turn on AI video generation

Optional, off by default. A generated clip costs money — an 8-minute video is
~96 five-second segments, so generating all of them is expensive. That is why
the default when you enable it is hybrid: the AI covers the first clips
(the hook, where retention is won) and Pexels fills the rest.
VIDEO_AI_MAX_CLIPS is the per-video cap and is never exceeded; generated
clips are cached, so re-running the same topic does not pay twice.

Configure it from the TUI (tube-assistant clips, or Video clips in the
menu): source, provider, API key, model and the cap. If the provider is down or
out of credit the pipeline falls back to Pexels instead of publishing a video
with no b-roll.

Architecture

youtube_ai_agent/       CLI and launcher (pip entry point)
moduli/
  cervello.py           Topic and script generation (LLM)
  audio.py              Voice synthesis (Edge TTS)
  asset.py              Pexels clip download + AI clip pool assembly
  video_ai.py           AI video generation (Replicate / fal.ai / Luma / OpenRouter)
  impostazioni_video.py Clip-source settings screen (wizard + TUI)
  montaggio.py          Video editing (FFmpeg / MoviePy)
  thumbnail.py          AI thumbnail (Pollinations.ai / HuggingFace + Pillow)
  pubblica.py           YouTube Data API v3 upload
  analytics.py          Channel performance reading
  strategia.py          Strategy adaptation from analytics
  telegram_handler.py   Telegram bot (live control)
  ai_client.py          Multi-provider AI client
agent.py                Main daemon (24/7 scheduler)
wizard.py               Configuration wizard

License

MIT — use, modify, and distribute freely.