- Home
- Skills
- Productivity
- Transcribe by shash.builds
Transcribe by shash.builds
Audio or video to transcript, SRT and VTT captions, and word timestamps with local Whisper. English, Hindi, Hinglish. Free, offline, no API key.
Free
Works with the AI Agents you already use
Transcribe by shash.builds
Example session with this skill installed
Transcribe reel.mp4 and make captions. It's in English, and the brand is spelled shash.builds.
- Read your context and instructions
- Compiled the transcribe by shash
Ran python transcribe.py reel.mp4 --model small --language en --prompt "shash.builds". Wrote reel.txt, reel.srt, reel.vtt, reel.json and reel.words.json, shows the transcript, says how many words were unsure, and offers the next step (cut-silence, video-build or repurpose).
Connects securely to your tools. The creator never sees your data.
About this skill
Audio or video in, transcript and captions out, on your own computer. Local Whisper (faster-whisper), English, Hindi and Hinglish. No API key, nothing uploaded.
Point it at an mp3, wav, m4a, mp4 or mov and you get five files: a plain transcript, SRT captions, WebVTT captions, the engine's segments as JSON, and a flat word list with every word's start and end time.
What it's good for
- Captions for reels, Shorts and YouTube videos, cut into short lines (42 characters by default) that read well on a phone
- Word timestamps for editing tools, silence cutting and burned-in word-by-word captions
- English, Hindi, and Hinglish written in Roman letters (one flag)
- Spelling names right: pass the names you expect and Whisper uses them
How it works
- Model choice:
small(default, fast, about 470 MB, downloaded once) ormedium(more accurate, better for Hindi, about 1.5 GB);baseandlarge-v3also work - Point
--model-dirat a faster-whisper model folder you already have to work offline - The same engine as the Transcriber by shash.builds skill, with toolbox defaults
Tested
On Linux with Python 3.13 and faster-whisper 1.2.1, a 7.8 second synthetic test voice transcribed with the small model in about 10 seconds on CPU: "Hello and welcome to shash.builds. Today we build a one-page website for a local plumber. It takes about seven days."
Free and MIT licensed. Part of the shash.builds Toolbox.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Productivity.
Runs on the CPU. The first run downloads the model once (small about 470 MB, medium about 1.5 GB), from Hugging Face until the verified shash.builds model packages are published. Whisper can mishear names, numbers and mixed languages; Hinglish mode runs but has only been tried on a synthetic test file. Captions are re-made, not edited, when the video is cut.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 13 days ago
- Free with an Agensi account
- Accounts
Hugging Face
- Software
Python 3.9+faster-whisperPyAVNumPyInternet access
- Optional
- FFmpeg
PyAV must be below version 19. Install them with pip install -r requirements.txt. Internet is needed once, to download the model. FFmpeg is recommended.
Permissions
Allowed Hosts
File Scopes
Runs on your own computer. The audio is never uploaded; only the model is downloaded, once.