1,200+ early adopters·50,000+ videos processed·92% execution success rate

Learn from any video.Execute every step.

AI-grade accuracy

WithVideo is an AI-powered learning and execution engine for engineers: it parses tutorial videos, distills the key steps, and runs them right inside your environment.

Install and try See examples

wv — ~/my-video-projectv0.1.0

$ wv learn https://b23.tv/BV1xxx4y7YYY

[wv] ▶ transcript

[wv] ✓ whisper [45/45] (78s)

[wv] ✓ transcript (92s)

[wv] ▶ vision

[wv] ✓ ocr [12/12] (98s)

[wv] ✓ vision (203s)

[wv] ✓ semantic (165s)

[wv] ✓ guide (14s)

[wv] ━━━━━━━━━━━━━━ 100% | total 484s

✓ run_completed

→ .withvideo/run-af3c/guide.md

→ .withvideo/run-af3c/semantic.json

→ .withvideo/run-af3c/code/

$ wv act .withvideo/run-af3c/guide.md

Three ways to start

Up in 30 seconds

Pick the path closest to your workflow and start using WithVideo now.

Fastest

pip install

Python 3.10+. One command installs the CLI. Use withvideo on local files or URLs right away.

pip install withvideo

See CLI usage →

Claude Code plugin

Inside Claude Code, call @withvideo and let Claude parse, plan, and execute every step for you.

claude /plugins add withvideo

Plugin docs →

MCP server

Any MCP-capable agent or IDE plugs in: Claude Desktop, Cursor, Windsurf, custom agents — all work.

npx -y withvideo-mcp

MCP setup →

Core capabilities

Video → knowledge → execution

Structured video parsing

Automatically extracts steps, commands, and core points from tutorial videos into a searchable knowledge index.

Straight into your environment

Not a notes dump — real execution inside your local shell, editor, and IDE.

Pro

Plugin ecosystem

YouTube, Bilibili, local files, LLM backends, executors — all swappable.

Multilingual output

Handles EN, ZH, JA, KO source videos. Outputs in Chinese or English, switch anytime.

FAQ

Does my video have to leave my machine?

No. The CLI processes everything locally by default. A hosted tier, when launched, will be opt-in.

Which video sources are supported?

Local files (mp4/mov/mkv), YouTube, Bilibili. Additional platforms via plugins.

Is execution safe — can it mess up my environment?

Every step is a previewable, approvable diff. Nothing runs until you explicitly accept.

How does it integrate with Claude Code and Cursor?

Use the Claude Code plugin or the MCP server — detailed setup in the docs.

How does pricing work?

Personal use is free. Pro adds hosted execution, team knowledge bases, and priority support, billed on a subscription.

WithVideo vs. just watching a YouTube tutorial — what's the difference?

Watching, you still have to type every command yourself and hope you don't miss a step. WithVideo extracts the steps and executes them in your local environment — not notes, actual results.

What types of tutorials work best?

CLI walkthroughs, framework bootstrapping, and deployment flows work best. Pure theory videos with no concrete steps are a poor fit — WithVideo is optimized for content with clear, executable actions.

Which LLM does it use under the hood? Can I swap it?

Defaults to Claude (claude-sonnet-4-6). Pass --llm to switch to OpenAI, Gemini, or any OpenAI-compatible local model (e.g. Ollama).