Rideo

Rideo helps agents turn instructional videos into detailed, illustrated lessons. It combines subtitles and timestamped frames with skills that guide the agent to preserve explanations, reasoning, worked examples, and practical steps in readable Markdown.

It works with existing agent hosts through a Codex plugin and a pi extension, backed by a shared TypeScript core and CLI.

What it does

• Reads local SRT/WebVTT subtitles or retrieves available platform captions with yt-dlp.
• Prepares timestamped evidence with independent survey and initial image budgets.
• Explores local intervals around concrete questions, returning color previews, before/after candidates, and observation gaps.
• Promotes selected timestamps to full-resolution evidence; final lessons have no image count limit.
• Exports selected illustrations as portable image assets with source references.
• Guides lesson writing with the rideo skill and offers an explicitly requested review through rideo-review.
• Answers readers' questions with rideo-qa and expands the relevant passage in the existing lesson.

The output is a Markdown lesson with accompanying images. Agent additions and corrections are marked separately from source material. Real ASR, media downloading, and clip extraction are not implemented yet.

Get started

Install the public npm package in pi:

pi install npm:rideo

For the standalone CLI, use npm install -g rideo. See the host integration guide for Codex installation, updates, and release archives. Distributed packages contain built JavaScript; installation does not require a TypeScript build.

Once installed, ask your agent:

Use the rideo skill to turn /work/lesson.mp4 and /work/lesson.vtt into a detailed illustrated lesson at /work/lesson-notes/tutorial.md. Preserve the explanations and worked examples.

While reading, ask for a clarification in the same manuscript:

Use rideo-qa on /work/lesson-notes/tutorial.md: I do not understand why we divide by the sample count in the loss section. Explain it with a small example and update that passage in the same first-person teaching voice.

Runtime requirements: Node.js 24+, FFmpeg/ffprobe 6+ for video preparation and new frames, and yt-dlp for platform subtitles. Reading and exporting existing evidence bundles requires neither the original media nor those external tools. The agent must support viewing images.

Develop locally

pnpm install
pnpm build
node dist/cli.js --help

Use Node.js 24+ and pnpm 12.4.1. Run pnpm package:hosts to build distributable Codex and pi packages.

Before committing, run:

pnpm check
pnpm test
pnpm test:hosts
pnpm test:integration

The host and media integration tests require FFmpeg/ffprobe. They use local fixtures and fake subtitle services, so no provider credentials or network access are required. test:hosts also builds and checks both distribution packages.

Documentation

• Tool reference — CLI commands and evidence workflow
• Host integrations — installation, updates, packaging, and distribution
• Authoring skill, reader Q&A skill, and review skill
• Project intent — goals and architectural constraints
• Frame selection and extraction — algorithm tradeoffs, decoding costs, and coverage limits