Supametas.AI is a platform that transforms unstructured data into structured formats suitable for use in large language models (LLMs) and retrieval-augmented generation (RAG) systems. The platform is designed to simplify data collection, construction, and preprocessing for industry-specific datasets, making it easier for companies to bypass complex data cleaning processes. Users can convert data from multiple sources such as APIs, URLs, local files, images, audio, and video into JSON and Markdown formats, which are then seamlessly integrated into LLM RAG knowledge bases.

Key Features:

Versatile Data Collection: Supports data ingestion from multiple sources, including APIs, web pages, local files (docx, pdf, txt, md, json), images (jpg, png), audio (mp3), and video (mov, mp4, mpv). Standardized Output: Extracts data into standard JSON and Markdown formats, ensuring compatibility with various LLM frameworks. LLM RAG Integration: Seamlessly integrates with LLM RAG knowledge bases, including OpenAI Storage and Dify Datasets, with API support for custom integrations. User-Friendly Interface: Offers a zero-threshold, out-of-the-box experience, enabling quick creation of industry datasets. Data Privacy: Provides options for both SaaS and private Docker deployment to address enterprise data privacy needs.

Use Cases:

Knowledge Base Creation: Rapidly build and maintain LLM knowledge bases with structured data extracted from diverse sources. Data Preprocessing: Streamline data preprocessing pipelines for LLM applications, reducing manual effort and improving data quality. Data Processing: Process digital human data for use in AI applications. Content Transformation: Transform raw data into desired content formats, boosting productivity and efficiency. Podcast/Video Data Integration: Convert podcast audio and video data into LLM knowledge bases.

TranscriptFetch is one API for getting text out of video and web content.

Send a URL from YouTube, TikTok, Instagram, X or Facebook and get back clean, timestamped text. Send any web page and get clean Markdown. One endpoint, one response shape, one API key.

The part that actually matters

Most short-form video has no caption track to download. TikTok's auto-captions are opt-in per upload, Instagram never publishes a downloadable track, and a large share of captions on both platforms are burned into the video frames where no parser can read them.

When there is no caption track, TranscriptFetch transcribes the audio instead. Same endpoint, same response, so your code never branches on which method produced the text.

What you get back

• A joined text field for feeding a model or a search index
• A segments array with per-cue start times and durations, so subtitles and jump-to-moment links are a formatting step rather than another integration
• Consistent output whether the text came from captions or speech recognition

Built for pipelines and agents

• MCP server so Claude, Cursor and other MCP clients can fetch transcripts as a tool mid-conversation
• Python and JavaScript SDKs
• Batch endpoint for up to 50 videos in a single call
• Channel, playlist and keyword-search endpoints for ingesting at scale

Pricing

100 free credits on signup, no card required. One credit per successful response. Failed, blocked and empty results are never charged, which matters on short-form video where a meaningful share of any batch is music with no speech in it.