laya-browser-agent: browser decisions on Laya, drop-in for jev-ultrafast tooling

# laya-browser-agent

**Multi-backend constrained decision support for Jev-like models.** This project can run
Laya through MLX or PyTorch, accept any synchronous `answer(state, questions)` duck backend,
or call an arbitrary System One-shaped HTTP endpoint. The recommended browser default is
the project's own open checkpoint, not a forced proprietary model.

**[English](README.md)** | [中文](README.zh-CN.md) | [日本語](README.ja.md) | [Español](README.es.md)

[](https://github.com/ChenneyZhuang/laya-browser-agent/actions/workflows/tests.yml/badge.svg)
[](LICENSE)
[](https://huggingface.co/ichenney/laya-browser-v32b)



A local, open-weight alternative to [TypeSafe Jev](https://docs.typesafe.ai) for the
browser-driving use case — running [Laya](https://github.com/NandhaKishorM/laya), the
open-source "System One" decision model, fully on your own machine. Works with
[browser-use/jev-ultrafast](https://github.com/browser-use/jev-ultrafast) via the same
wire format, and speaks TypeSafe's `/v1/systemone` dialect, so existing Jev tooling
points at it by changing one base URL.

A decision model answers typed questions about a state and returns probability
estimates over the supplied options. It does not generate instructions, but it
can still be wrong or confidently wrong; validation, outcome checks, and human
confirmation remain necessary. That makes it a useful *decision* component for
a browser agent: give it a numbered table of controls and constrain which
operation and element it may propose.

This project wires those models into that role, locally, for whatever agent you
already use. The official [Laya base model](https://huggingface.co/convaiinnovations/laya)
is distinct from the upstream browser fine-tune
([`cklxx/laya-browser`](https://huggingface.co/cklxx/laya-browser)).