yt-mem-ai 0.2.0

Local-first YouTube AI CLI: download, transcribe (captions -> whisper), embed in LanceDB, discover subscriptions, and summarize/highlight/Q&A.

yt-mem-ai — YouTube AI CLI

Install

# zero-install run (recommended)
uvx yt-mem-ai --help

# or bootstrap uv + warm the cache
curl -LsSf https://raw.githubusercontent.com/dasein108/yt-mem-ai/main/install.sh | sh

The desktop UI lives in a separate repo: yt-mem-ai-desktop. It depends on this yt-mem-ai package and runs its own local REST API.

Download YouTube audio, transcribe (captions → faster-whisper fallback), store everything in an embedded LanceDB with per-chunk embeddings, discover subscription uploads, and search your library semantically. Summaries, highlights, and Q&A are produced by Claude Code skills, not an API.

Setup

uv sync --extra dev
cp .env.example .env   # fill WEBSHARE_* + YT_COOKIES_BROWSER; pick embedding backend

Config (.env): YT_STORE_PATH (LanceDB dir), YT_EMBEDDING_BACKEND=local|openai, YT_EMBEDDING_MODEL, YT_CHUNK_TARGET_S, OPENAI_API_KEY (openai backend), WEBSHARE_PROXY_*, YT_COOKIES_BROWSER.

Proxy / VLESS: YT_USE_WEBSHARE defaults off. If you already run a system-level proxy/VPN (VLESS/Xray etc.), leave it off — traffic rides that tunnel. Stacking the Webshare proxy on top breaks the authenticated subscription feed (its CONNECT tunnel returns 405). Only set YT_USE_WEBSHARE=true if you have no other proxy and YouTube rate-limits your raw IP. Discover tuning: YT_DISCOVER_FEED_LIMIT (newest-N cap, default 60), YT_DISCOVER_OVERLAP_S (incremental overlap, default 3600), YT_DISCOVER_TIMEOUT_S.

Commands

yt-ai fetch <url>            # download + transcribe + embed + store one video
yt-ai fetch <url> --captions-only  # captions only: no audio download / no whisper (fails if none)
yt-ai transcript <url>       # same pipeline
yt-ai discover               # new subscription uploads (--after/--deep/--min-duration/--json); incremental by default
yt-ai fetch-pending          # batch-fetch pending 'discovered' videos (since --since, default today; --limit)
yt-ai list                   # list stored videos (--status/--since/--json)
yt-ai show <video_id>        # metadata + transcript (--json)
yt-ai status                 # counts by status
yt-ai search "<query>"       # semantic search (--hybrid/--fts/--vector, -k N)
yt-ai save-summary <id> "<summary>" --highlights '<json>' --qa '<json>'  # persist a summary (used by skills)
yt-ai like <video_id>        # mark liked (feeds recommendations)
yt-ai dislike <video_id>     # mark disliked
yt-ai recommend              # rank your unrated fetched videos by taste (--limit/--json)
yt-ai compile                 # deep-linked highlights doc, budget-bounded (--since/--max-minutes/--json/--out)
yt-ai supercut                 # video reel of highlights, re-downloaded + labeled (--since/--max-minutes/--out/--keep-clips)
yt-ai frame <video_id> --at <ts>  # still frame at a timestamp (seconds or H:M:S) → frames/<id>_<s>s.png

Rate & recommend

Like/dislike videos you've fetched (yt-ai like <id> / dislike <id>), then yt-ai recommend ranks your other fetched-but-unrated videos by similarity to what you liked (minus what you disliked), using their transcript embeddings. Before you've liked anything, it falls back to most-recently-published.

Daily routine

yt-ai discover               # find new subscription uploads → 'discovered'
yt-ai fetch-pending          # download+transcribe+embed today's batch (robust, skips failures)
# then in Claude Code:
/yt process subscriptions    # per-video summaries + digests/YYYY-MM-DD.md
yt-ai compile                # compile the day's highlights into a deep-linked markdown you can click into

yt-ai discover is incremental: it pulls the newest feed entries (one flat call, capped by YT_DISCOVER_FEED_LIMIT), stamps each with an approximate timestamp (yt-dlp youtubetab:approximate_date), and keeps only those newer than the last run's stored high-water mark minus a 1h overlap (YT_DISCOVER_OVERLAP_S) — so hour-rounded dates never miss a boundary video, and already-processed videos (is_seen) are filtered out. Pass --after YYYY-MM-DD to override the cutoff manually. Full per-video metadata (description, tags, exact time) is fetched later at ingest, not during discover.

yt-ai compile builds compilations/<DATE>.md: each highlight from the day's summarized videos becomes a deep link (watch?v=ID&t=<start>s) that jumps straight to its moment, newest-video-first and budget-bounded by --max-minutes (default 20). Fast — no downloading, just the same compile_highlights selection rendered as markdown.

yt-ai supercut renders that same highlight selection as an actual video reel instead of a doc: it re-downloads each highlight's section (720p, yt-dlp --download-sections), burns a label onto each clip (title/timestamp), and concats them into one mp4 — so it needs network access and a local ffmpeg, and is much slower than compile. Output is supercuts/<since-or-today>.mp4 plus a sidecar supercuts/<...>.mp4.refs.md listing each rendered clip's source link (and any clips skipped because their download/render failed). Use compile for a quick clickable digest; use supercut when you want a shareable video.

Single video on demand: yt-ai fetch <url> then the /yt skill (summarize / highlights / qa / presentation).

Logging

The CLI writes structured JSON events to logs/common.jsonl (via obs.log_event/blog) — one object per line, {ts, source, level, event, msg, ...ctx}. Override the path with YT_LOG_FILE; it's gitignored. Inspect with jq:

jq -c 'select(.level=="error")' logs/common.jsonl   # every error
tail -f logs/common.jsonl | jq -c '{ts,event,msg}'  # live tail, compact

Tests

uv run pytest -q                       # offline unit tests (fake embedder)
YT_RUN_INTEGRATION=1 uv run pytest -q  # + real sentence-transformers integration