@bat-ai

Personal Dropbox / WeTransfer — self-hosted file share with shareable preview links. Upload any file (video, image, PDF, archive, anything up to 10GB) and get a short URL with inline preview + download. TRIGGER when the user says "redok this", "redok <file>", "upload to redok", "share via redok", "get me a link for <file>", "send <file> to <person>", "drop this on redok", "redok up", "make a redok link", or any request to share a local file as a public URL. Bring-your-own-server — point at any redok-protocol endpoint with your token.

Install Demonic Desktop, an Electron AI workspace app powered by OpenRouter. TRIGGER when the user says "install demonic desktop", "set up demonic", or asks to install the demonic chat app. Builds and installs a signed .app to /Applications on macOS.

Encoding/decoding swiss army for any text. Encode and decode base64 (standard, urlsafe, no-padding), base32, base16/hex, URL percent-encoding, HTML entities, ROT13, Caesar shift, binary, JWT decode (no verify), unicode escape. Auto-detect mode that guesses the encoding of an input string. Pure stdlib. Use when the user asks to base64 encode / decode, "url-encode this", "what's this hex", JWT decode, "decode this string", "what encoding is this", or any small encoding/decoding task.

Generate strong passwords + passphrases in multiple modes. Random ASCII (with tunable charset), memorable diceware passphrase (EFF wordlist, configurable separator + word count), pronounceable, hex, alphanumeric-only, PIN. Optional `--copy` puts it on the macOS clipboard, `--qty N` generates a batch, `--strength` shows entropy bits. Pure stdlib + bundled EFF short wordlist. Use when the user asks for a password, passphrase, "generate a strong password", "diceware", "random pin", or needs to produce credentials for an account.

Terminal countdown timer + Pomodoro mode with macOS notifications and optional sound. Accepts natural durations like `25m`, `1h30m`, `45s`. Lock-screen-safe (uses osascript notify), optional `--say "TIME"` voice announcement, optional `--terminal-bell` audio cue. Pomodoro mode runs alternating focus/break cycles with progress bar in terminal. Pure stdlib + macOS native. Use when the user asks for a timer, countdown, pomodoro, focus session, "remind me in N minutes", egg timer, or stopwatch.

PDF swiss-army knife. Merge multiple PDFs, split into pages or page ranges, rotate, extract specific pages, encrypt with password, decrypt (if you have the password), compress (via ghostscript when available), and extract text. One CLI with subcommands — `merge`, `split`, `rotate`, `extract`, `lock`, `unlock`, `compress`, `text`. Pure pypdf for most ops. Use when the user asks to "merge PDFs", "split a PDF", "extract pages", "lock a PDF with a password", "compress this PDF", or has any PDF manipulation task.

Encrypt and decrypt .env files with a password. Drop your real .env, get an .env.vault file safe to commit. Decrypt on a teammate's machine with the same password. AES-256-GCM via the cryptography library — strong + standard. Optional --rotate to change the password without exposing plaintext. Use when the user asks to "encrypt my .env", "share secrets safely", "vault my env", or doesn't want to use a SaaS like dotenv-vault.com.

Generate styled QR codes for URL, wifi password (auto-format), vCard, plain text, or arbitrary payloads. Custom foreground/background colors, optional center logo overlay (auto-resized + framed), optional rounded modules for modern aesthetic. Outputs PNG or SVG. Pure Python via qrcode + Pillow. Use when the user asks for a QR code, wifi QR, contact QR, "scan-to-pay", or wants a stylized QR for a poster / cover art.

Distill a folder of past prompts (or a saved conversation export) into reusable templates with `{placeholder}` variables. Heuristic detector — no LLM call needed. Finds high-frequency phrases, common openers, recurring task structures, and emits a `prompts.yaml` of named templates plus a frequency report. Pure Python stdlib + regex. Use when the user has a graveyard of past prompts and wants to extract a reusable library, or to find what they keep typing over and over.

Generate professional invoice PDFs from a JSON spec or CLI args. Includes header (logo + business info), bill-to block, itemized line table with quantity/rate/total, subtotal/tax/grand-total summary, due date, payment instructions, and optional notes. Built-in clean editorial style. Pure Python via reportlab — no LaTeX, no Office. Use when the user asks for an invoice PDF, billing statement, freelance invoice, or wants a one-shot "make me an invoice for $X" tool.

Translate cron expressions to plain English and show the next N fire times. Supports standard 5-field crontab and limited 6-field (with seconds). Optional reverse mode — natural-language phrasing → cron expression. Pure Python — uses croniter for next-fire times, hand-rolled English description. Use when the user asks "what does this cron run", "explain this crontab", "next runs of X", "convert daily 9am to cron", or is debugging a scheduled task.

Generate procedural desktop wallpapers — pure Python + PIL, no AI. Pick a color or list of hex codes, choose a style (gradient, mesh-blur, noise-grain, ascii-stripes, scanlines), optionally add a centerpiece (circle/ring/triangle). Output is a 5K (5120x2880) PNG ready to drop into System Settings → Wallpaper. Pairs cleanly with palette-rip — extract a palette from a reference image, then generate wallpapers in that palette. Use when the user asks for a wallpaper, desktop background, lock screen, or wants minimalist abstract art they can use as a bg.

One source PNG/SVG → complete favicon + app-icon set. Outputs favicon.ico (multi-size: 16, 32, 48), apple-touch-icon-180.png, android-chrome-192.png, android-chrome-512.png, mstile-150.png, plus a site.webmanifest and a copy-pasteable HTML <head> snippet. Optional --circle masks the icon into a circle, --trim removes alpha padding, --bg fills transparent areas with a hex color. Use when the user asks for favicons, app icons, icon set, "make icons for my site", or has a logo and needs every web/app icon size.

Markdown → polished slide deck. Splits on `---` separators, each section becomes a slide. Outputs a single self-contained HTML file with arrow-key navigation, OR a PDF via headless Chrome / wkhtmltopdf. Built-in dark editorial theme with serif headlines, monospace body, optional accent color. No extra deps beyond Python stdlib + the markdown library. Use when the user asks for a presentation, slide deck, pitch deck, talk slides, or wants to "turn this README into slides".

Test, explain, and sanity-check Python/PCRE regex patterns. Pass a pattern and test strings, get all matches with capture groups + a per-token explanation of the regex itself (anchors, quantifiers, character classes, groups). Pure Python stdlib. Use when the user asks "explain this regex", "test this regex", "build a regex for X", "what does this pattern match", or wants to debug a non-matching expression.

Generate social-share card images from a title + subtitle (and optional background image). Three formats — og (1200x630 Open Graph / Twitter), square (1080x1080 IG grid), story (1080x1920 IG/TikTok story). Polished editorial typography defaults, dark-mode aesthetic, optional gradient overlay over background images. Pure PIL. Use when the user asks for an OG card, share image, link preview, IG grid post, story image, or release announcement card.

Inspect any video file's metadata — resolution, codec, fps, bitrate, duration, audio details, color space, encoder, container tags, GPS coords, creation date, file size. Pretty single-file view OR table view for a folder. Optional --json for machine-readable output. Pure ffprobe wrapper. Use when the user asks "what's the resolution / fps / bitrate of this", "inspect this video", "ffprobe", "metadata of this clip", or wants a quick quality audit before processing.

Spin up a tiny local server that exposes your macOS clipboard as a plain-text URL. Useful for sharing code/text quickly with another device on the same network or via tunnel (cloudflared, ngrok). Optional history mode (last N clipboards), optional write-back mode (PUT to set clipboard). Pure stdlib http.server. Use when the user asks to "share my clipboard", "expose pasteboard", "quick paste server", or needs a fast way to get text from their Mac to another device without sending a message.

Convert any image to emoji art — output PNG (or text grid) where each cell is the color-closest emoji from a curated palette. Tunable cell size and emoji set (food, faces, nature, geometric). Pure PIL, no ML. Use when the user asks for emoji art, "make this image out of emojis", emoji mosaic, or wants a fun social-media-ready stylization of a photo.

Animated logo sting / intro generator. Drop in a logo PNG, pick a preset (fade-rise, glitch-stutter, neon-flicker, scale-bounce, spotlight-sweep, ink-bleed), get a 2-4s MP4 intro for music videos, releases, or YouTube channel openers. Pure ffmpeg + PIL. Optional sound effect overlay. Use when the user asks for a logo animation, intro sting, channel opener, brand intro, or wants their logo to "drop with motion" before a video.

Build a polished moodboard collage from a folder of reference images. Smart non-uniform packing layout (mix of large hero + small accent cells), optional dominant-color rail along the bottom and caption strip at the top. Outputs one print-ready PNG you can post or use as a video reference. Distinct from gif-board (uniform contact sheet) — mood-board is curated, not exhaustive. Use when the user asks for a mood board, reference collage, visual brief, art direction sheet, inspiration sheet, or "build a board for this project".

Record the macOS screen to MP4 with one command. Full screen, custom region, or any connected display. Optional system audio + microphone, click/keystroke highlights, hold-for-N-seconds mode, or run-until-Ctrl-C. Pure ffmpeg + AVFoundation, no GUI app, no driver install. Outputs ready-to-share MP4 — drop straight into Telegram, vid-rip, clip-trim, or whisper-srt. Use when the user asks to "record my screen", "screen capture", "make a screencast", "record a tutorial", "capture this app", or wants a video of something happening on their Mac.

Generate stylized + corrupted Unicode text variants from any input string. 14 styles — zalgo (combining-mark chaos), fullwidth, smallcaps, upside-down, struck-through, double-struck, fraktur (gothic), bold/italic mathematical, monospace, leet, vaporwave, cursed, mirror, censored block. Outputs all styles at once or just the one you pick. Pure Python, no deps. Use when the user asks for "weird text", "fancy text", "zalgo text", "corrupted typography", "social caption variants", "make this text scary", "fonts for IG bio", or wants typographic variants for cover art / video text overlays.

Auto-detect and blur faces in images or videos. Privacy-protect footage with one command — gaussian blur, pixelation, or solid emoji block. Uses OpenCV Haar cascades (fast, no ML model download). Folder mode processes everything in parallel. Use when the user asks to "blur the faces", "anonymize this video", "censor faces", "pixelate the people", or needs to scrub a clip of identifiable subjects before posting.

Change the tempo (BPM) of any audio file without altering its pitch — or change the pitch without altering the tempo — or both at once. Auto-detects source BPM via librosa, then time-stretches to your target. Optional pitch shift in semitones for key matching. Drop-in for sample/loop matching, DJ prep, "make this fit my beat", or repitching a vocal up/down. Use when the user asks to "stretch this to 140 bpm", "match the tempo", "change pitch without changing speed", "warp this sample", or "make this fit my beat".

Analyze any audio file and extract its chord progression with timestamps. Outputs a labeled chord chart (text + JSON), the detected key, tempo, and Roman numeral analysis (I — IV — V — vi etc) so you can see the function of each chord. Uses librosa chroma + chord-template matching, plus optional MIDI export of the chord roots. Distinct from midi-rip (full polyphonic transcription) — chord-rip gives you the harmonic sketch, not every note. Use when the user asks "what chords are in this", "get the chord progression", "analyze the harmony", "transcribe to chords", or wants a chord chart for a sample/song they want to flip.

Extract stills from any video — at intervals, evenly-spaced, at scene changes, or just I-frames. Optional sprite-sheet output (one big PNG with N frames in a grid). Pure ffmpeg, no ML deps. Distinct from gif-board (folder → contact sheet) — frame-grab is one video → many frames. Use when the user asks for thumbnails, screenshots from a video, every-Nth-second frames, scene-change detection, sprite sheet, key-frame extraction, or stills for OCR / clip-search / cover-art workflows.

Concatenate a folder of video clips into one MP4 with optional crossfade, fade-to-black, or hard-cut transitions. Auto-normalizes resolution / framerate / SAR so mismatched source clips merge cleanly. Order by filename, mtime, or shuffle. Optional uniform clip duration (--per-clip) for rhythmic edits. Use when the user has a folder of clips and asks to "stitch", "concat", "join", "merge these clips", "build a reel from this folder", or wants a quick montage.

Generate .srt / .vtt / .json subtitles from any audio or video using OpenAI Whisper (local, no API). Auto-detects language, supports translation to English, outputs word-level timestamps when needed. Optionally burns the subtitles directly into the video as a styled hardcoded track. Distinct from lyric-engine (music videos) — whisper-srt targets spoken content: interviews, podcasts, vlogs, lectures, voice memos. Use when the user asks for subtitles, captions, transcribe to SRT, "add captions", "burn subs", or wants a transcript file.

Auto-trim dead air from clips. Detects leading/trailing silence in audio (or audio track of a video) AND leading/trailing black frames in video, then trims both. Pure ffmpeg silencedetect + blackdetect — no ML deps, no re-encode for video when possible. Folder mode trims every file in a directory in parallel. Use when the user has clips with dead intros, fade-up black frames, silent leading air, render-padding at start/end, or just wants every file in a folder cleaned up.

Auto-fill MP3/M4A/FLAC ID3 metadata using Shazam (via shazamio). Detects the song from each file, then writes title, artist, album, year, genre, track number, and embeds high-res cover art. Folder-aware: tags every audio file in a directory in one pass. Skip already-tagged files with --skip-tagged. Use when the user has a folder of untagged audio dumps, asks to "tag these mp3s", "fill in metadata", "add cover art to these tracks", "fix my music library", or wants to clean up a rip/dump folder for Apple Music/Spotify uploads.

Animated typography overlays for video. Drop a title card, drop announcement, callout, intro text, or release tag onto any clip — or render text alone on black. Built-in motion presets like kanye-zoom (crash-in), drain-stutter (frame glitch), brutalist (hard cut mono caps), rise-fade (cinema fade), noir-typewriter (letter-by-letter), tape-line (VHS scanlines). Pure ffmpeg drawtext, no ML deps. Use when the user asks for a title card, intro text, animated text, drop tag, callout, "put text on this clip", "make a release card", or wants typography that's not full lyric video.

Turn a folder of random audio files into a standardized, releasable sample pack. Auto-detects BPM and key (via librosa), renames files with `{bpm}bpm_{key}_{name}.wav`, converts everything to 24-bit/44.1kHz WAV, sorts into one-shot vs loop subfolders by duration, builds a JSON manifest, and writes a README. Drop-in producer pipeline for FL Studio / Ableton libraries. Use when the user asks to "make a sample pack", "organize these samples", "tag with BPM and key", "build a kit from this folder", or wants to turn a dump folder into a usable producer asset.

Smart yt-dlp wrapper. Download any video (YouTube, TikTok, IG Reels, Twitter/X, SoundCloud, etc) with auto-best-quality format picking. Optional reframe to 9:16 (Reel/Short), 1:1 (square), or 16:9 (landscape) via centered crop. Audio-only mode for sampling. Strips watermarks via cropping when possible. Use when the user pastes a video URL and wants it downloaded, asks to "rip this", "download this video", "save this Reel", "get the audio from", or wants footage prepped for a 9:16 edit.

Build a contact sheet (grid of thumbnails with filenames) from a folder of GIFs, MP4s, MOVs, or WebMs. Static PNG mode and animated GIF mode (all cells loop simultaneously). Dark monospace aesthetic, perfect for browsing clip libraries before a music-video edit. Use when the user asks for a contact sheet, thumbnail grid, clip overview, mood board from videos, "show me all the clips in this folder", or wants to scan a B-roll library at a glance.

Extract a designer-ready color palette from any image or video. Outputs .hex (plain list), .gpl (GIMP/Inkscape), .ase (Adobe Swatch Exchange), CSS variables, and a labeled PNG swatch sheet. K-means clustering picks N dominant colors; for videos, samples N frames and merges. Use when the user asks for a color palette, design tokens, swatch sheet, brand colors, dominant colors, "what colors are in this", or wants to turn a reference image into a reusable palette for a UI/design system. Distinct from lut-snap (3D LUTs for grading) — palette-rip produces design tokens.

Identify any song from an audio file, video file, or YouTube/SoundCloud URL using Shazam's public API (via shazamio — no API key needed). Returns title, artist, album, year, Apple Music + Spotify + YouTube links, and cover art URL. Multi-section mode chops long files into chunks to catch mashups, medleys, or DJ sets. Use when the user asks "what song is this", "shazam this", "ID this track", "what's playing", or sends an audio/video clip and wants to know the source song.

One-shot background remover. Turn any image (or folder of images) into transparent PNG stickers using rembg's U2Net model. Outputs clean alpha cutouts ready to be layered as VFX assets in cinema-x-gif, kromvision-gif-flips, or any compositing workflow. Use when the user asks for "rip the background", "make this transparent", "cutout the subject", "sticker pack", "alpha channel cutout", or wants to extract subjects from photos for sample-flip / VFX overlays.