Local AI models connected like Unix pipes. Drop a file, watch it flow.
AI models are powerful, but they're stuck in silos. A vision model in one app. An LLM in another. A transcription tool somewhere else. They don't talk to each other.
MachineFabric connects them. Drop a file, watch it flow through extractors, vision models, LLMs—whatever chain makes sense. The models don't know about each other. MachineFabric handles the wiring.
Unix solved this fifty years ago: small tools that do one thing, connected by pipes. The output of one becomes the input of the next.
MachineFabric treats AI models the same way. Each model is a black-box command. A vision model takes an image, outputs a description. An LLM takes text, outputs text. MachineFabric chains them together.
Plugins advertise what they can do using capability URNs. What inputs they accept, what outputs they produce. MachineFabric matches capabilities to find chains automatically.
The protocol is minimal: stdin/stdout with CBOR frames. Any language that can read and write bytes can be a plugin. The community writes plugins without coordinating with us.
cap:in="media:application/pdf";op=extract
cap:in="media:image/png";op=describe
cap:in="media:text/plain";op=translate;lang=fr
MachineFabric runs AI models on your Mac using MLX and Metal. Vision models, LLMs, transcription—all local.
No API keys. No usage limits. No sending your files to a server. Download a model once, run it forever. Works offline.
Plugins are standalone binaries. Any language that can read stdin and write stdout works. We have reference implementations in Rust, Swift, Go, and Python.
Someone wraps a new open-source model? Package it as a plugin. A better PDF extractor drops? Wrap it. The ecosystem grows without bottlenecks.
MachineFabric is a native Mac app. SwiftUI, AppKit, Metal for GPU acceleration. It feels like part of the system because it is.
Drag files from Finder. Results appear in a native window. No web views, no embedded browsers, no "this page is unresponsive" dialogs.
No cloud uploads. No "processing on our servers." AI models run locally. Everything is stored locally. Unplug your internet—MachineFabric keeps working.
We're launching with plugins for PDFs, EPUBs, images, and text files. Vision models for image understanding. LLMs for text processing.
Working now: PDF extraction, image description, MLX inference, the plugin protocol, native macOS UI.
Coming: More models, more file types, audio transcription, video processing—whatever the community builds.
If you want to try MachineFabric, put your email in. We'll send you a download link when it's ready.