We are very excited to tell you that yzma now runs local AI models right in your browser.

Version 1.26 turns llama.cpp into a WebAssembly module, and a Go program compiled with TinyGo drives it. The JavaScript glue picks the best build that the browser can run, which is WebGPU, more than one thread, or a single thread. Text, embeddings, and images all work, along with chat templates and tool calling. The calls have the same names and the same order as the ones you already use on Linux, macOS, and Windows, so the same code goes to a new place.

This release also verifies what it installs. Install compares the SHA-256 of every file against a published manifest, a version can pin that manifest, and the new yzma verify command checks an installation later.

Give it a try at https://yzma.ai/try/.

Code repository is at https://github.com/hybridgroup/yzma.