TTS_PROVIDER=openai or TTS_PROVIDER=none. See Setup.
1
Start a project
create-vfrey for you. npx create-vfrey my-channel does the same.
This makes the folder my-channel, installs vfrey in it, and runs vfrey doctor. Each failed check prints a Fix: line. The project already holds what every agent needs: AGENTS.md, CLAUDE.md, GEMINI.md, and the vfrey skill. See Projects.2
Go into the folder
3
Open it in your agent
Start your agent in this folder:
claude, codex, grok, gemini, or open the folder in Cursor. Nothing else to install. The agent reads the folder’s instructions and finds the skill there. See Your agent.4
Ask for a video
Say: “Make me a short video about how octopuses have three hearts, and show it to me.”The agent writes the script, designs a scene for the idea, renders the video into Without
videos/ under a readable name like videos/2026-10-26 How octopuses have three hearts.mp4, and shows it to you. The transcript sits beside it as .txt. The first render downloads Chrome Headless Shell (about 170 MB) once, so it takes longer.You can run it by hand too:--script, vfrey writes a basic placeholder script. An agent writes its own. See Your agent.5
Look at it
npx vfrey videos lists your videos with their paths.During the beta use
@beta: npm create vfrey@beta my-channel. The @latest form works once the beta is promoted.A fresh install has no platform key. The summary says “Replies skipped: no platform keys”. Upload is skipped (
Upload skipped: no key for youtube) and the video stays on your machine until you add your own keys. npx vfrey push refuses for now. Publishing happens inside tick once keys are set.tick makes nothing, it prints why. See Troubleshooting. To keep going on the series cadence, run npx vfrey worker.