Download audio from video URLs at maximum speed. Built by Augent with speed optimizations (aria2c multi-connection, concurrent fragments). Downloads audio ONLY - never video. Supports YouTube, Vimeo, TikTok, Twitter, SoundCloud, and 1000+ sites. IMPORTANT: When a user pastes ANY URL and asks what was said, what happened, or wants the content — use this tool to download, then transcribe_audio to get the text. This is the correct path for tweets, videos, podcasts, and any link with audio/video content. Do NOT use WebFetch for media URLs.

Transcribe an audio file and return the full text with timestamps. Useful when you need the complete transcription rather than searching for specific keywords.

Search audio files for keywords and return timestamped matches with context snippets. Useful for finding specific moments in podcasts, interviews, lectures, or any audio content.

Search audio by meaning, not just keywords.

Take notes from a URL. Downloads audio, transcribes, and saves .md to Desktop. This single tool handles the entire pipeline — download, transcribe, and save — when the user asks for notes, summaries, highlights, takeaways, eye-candy, quiz, or any formatted content from a video/audio URL. Returns audio_path for follow-up tools (chapters, search). Also used to SAVE formatted notes: call with save_content to write notes to the file from the previous take_notes call (no url needed).

Auto-detect topic chapters in audio.

Search multiple audio files for keywords in parallel. Ideal for processing podcast libraries, interview collections, or any batch of audio files. Returns aggregated results with file paths.

Convert text to natural speech audio using Kokoro TTS. Saves an MP3 file. Runs in background — returns a job_id immediately. Call again with job_id to check status. Pass text for raw TTS, or file_path to read a notes file (strips markdown, skips metadata, embeds audio player).

Find where one keyword appears near another keyword in audio. Useful for finding contextual discussions, e.g., 'startup' near 'funding'.

Identify who speaks when in audio.

List media files in a directory.

List all stored transcriptions with their titles, durations, dates, and file paths to markdown files. Useful for browsing what has already been transcribed.

View transcription memory statistics including number of stored files and total duration.

Clear the transcription memory to free disk space.

Separate audio into stems (vocals, drums, bass, other) using Meta's Demucs v4. Isolates vocals from music, background noise, and other sounds. Use this before transcription when audio has music, intros, or heavy background noise for dramatically cleaner results.

Search across ALL stored transcriptions. No audio_path needed, queries everything in memory. Default mode is 'keyword' (literal match). Use 'semantic' mode for meaning-based search.

Export a video clip from a URL for a specific time range. Downloads only the requested segment — not the full video. Perfect for extracting moments around keyword matches. Supports YouTube and 1000+ sites.

Add, remove, or list tags on a transcription. Tags are broad topic categories (e.g. 'AI', 'Health', 'Music') that help organize and filter memories in the Web UI.

Export MP4 clips of specific moments. Two modes: auto (AI picks top moments by quotability and insight density) or focused (find moments matching a specific topic, person, or concept). Returns timestamps and text for each highlight, the calling agent decides which to export as clips.

Rebuild Obsidian graph view data for all transcriptions. Migrates .md files to YAML frontmatter format, computes [[wikilinks]] between semantically related transcriptions, and generates MOC (Map of Content) hub files for tag clusters. Safe to run repeatedly. Run this once to upgrade existing memory for Obsidian.

ONLY call this tool when the user explicitly asks for visual context, screenshots, or frames. Do NOT call it alongside take_notes or download_audio unless the user specifically requests visuals. Extract visual context from a video at moments that matter. Four modes: (1) Query mode (default): describe what you need visual context for and the tool finds those moments in the transcript, then extracts frames. (2) Auto mode: autonomously detects moments where the speaker implies visual content (UI actions, screen recordings, demonstrations). (3) Manual mode: extract frames at specific timestamps. (4) Assist mode: analyzes the transcript for visual gaps and returns time ranges where the user should provide their own screenshots (ideal for talking-head videos where the speaker describes a UI but doesn't show it). Frames are stored in augent memory alongside the transcription and embedded in the .md file as Obsidian wikilinks.

Download or live-record a Twitter/X Space. Three modes: (1) Pass url to start a download (returns recording_id), (2) Pass recording_id to check status, (3) Pass recording_id + stop=true to stop a live recording.