Local inference
Model setup, memory constraints, and generation state belong in the product experience. Supported speech workflows run locally after setup.
Local AI · macOS / Independent product
Murmur
Turn text, PDFs, and EPUBs into audio with local models and a native Apple Silicon workspace.
Visit the product ↗SwiftUI / MLX / Local speech / MCP
The engineering challenge
From capability to product.
A model that generates speech is only the beginning. People also need model setup, document handling, long-running generation, and a release they can install and update.
Decisions that matter
One engine, several entry points
The native studio, command-line tools, and MCP interface expose the same core capability for different workflows.
A complete release
Licensing, signing, notarization, and updates are part of the product, alongside the interface and model integration.
What this demonstrates
End-to-end product engineering: connecting model capabilities to usable interfaces, managing processing state and failures, and preparing a release. The platform and workflow here are specific to this product; a business integration would have its own data, permissions, and acceptance checks.
My role: product design, engineering, model integration, and release work. This is an independent product, not a client engagement.
Building a similar workflow? Let’s talk ↗