Onde Inference

The Forward Pass.

Writing about local inference, Apple Silicon, and why the cloud was just a stopgap.

Engineering

Fine-Tune a Language Model on Your Mac

Onde CLI now lets you LoRA fine-tune a model, merge the adapter, export to GGUF, and upload to HuggingFace — all from a terminal UI running on Apple silicon. No cloud GPUs. No Python. No notebooks.

Privacy

Privacy as a Feature, Not a Compliance Check

There are two types of privacy in software today. The one written by lawyers, and the one built by engineers. Here is why the difference matters.

Company

What Onde Is

Onde lets your iOS or macOS app run a language model locally, on the device, without calling any server. This is what that means, what it solves, and where it fits.

Performance

The Latency War

We gave generative AI a free pass on speed for two years because it felt like magic. That honeymoon is over. People just want their apps to be fast again.

Architecture

The Death of the Cloud

We've spent the last decade treating our phones like dumb terminals for AWS. But the hardware caught up. The cloud was just a stopgap.