v0.4.0 — chapter outlines for long videos

Long videos get an outline instead of a transcript dump. Chapters come from YouTube or description timestamps. No new runtime dependencies.

New commands and options:

• info VIDEO: metadata and chapters, without the transcript
• search --match phrase: require query words to appear next to each other
• add/get/show JSON includes duration_s and chapters
• list --json includes uploaded_at and ingested_at
• Markdown exports include a chapter table of contents
• Human-mode warning when printing a long transcript without a time window

Existing v0.3 libraries migrate in place (schema 2) and gain chapters from saved descriptions. After migration, v0.3 will refuse the database.

Verified: 61 tests across macOS/Linux and Python 3.9/3.13.

Caption-less videos remain unsupported.

v0.3.0 — simpler setup, more useful transcript retrieval

A clearer, copy-first website and a more useful CLI, without new runtime dependencies.

New commands and options:

• get URL: save if needed, then read the transcript (supports time windows and JSON).
• get/show --plain: transcript text without metadata or timestamps.
• search --context 15: include surrounding captions without losing the original match.
• search --match any: broaden keyword searches.
• list QUERY --limit 20 --offset 0: filter by title/channel/ID and paginate.
• Explicit cached-language mismatch errors and per-command help.

The website now has copy buttons for every example, static macOS/Linux setup, and a plain-text llms.txt agent guide.

Verified: 53 tests across macOS/Linux and Python 3.9/3.13, live YouTube ingestion, seven clipboard targets, six screen widths, and automated accessibility checks.

Uses the same database schema as v0.2. Caption-less videos remain unsupported.

v0.2.0 — reliable transcripts for coding agents

Save available YouTube captions locally and retrieve timestamped passages from your coding agent.

• Preserves raw captions, technical notation, multiline text, and repeated speech.
• Passage-level SQLite FTS5 search and bounded reads with --from / --to.
• Idempotent ingestion, atomic markdown exports, consistent JSON output.
• Pinned installer and Pi/Codex/Claude Code skill with explicit cookie consent.
• Automatic v0.1 database backup and migration; legacy timing is marked approximate.

Validation: 34 offline tests passing across macOS/Linux and Python 3.9/3.13, plus live manual-caption and auto-caption ingestion checks.

See the README for installation. Requires Python 3.9+ with FTS5 and yt-dlp. Caption-less videos remain unsupported.