Vons

Plan globally. Decide locally. Keep the host in control.

Compact local decision models for frontier-agent workflows.

Kwangseob Ahn · INLEVEL9 / SEJONG UNIV. · oswarld@inlevel9.com

Source preview

This repository contains research source and documentation. It does not contain pretrained model weights, tokenizer assets or datasets. Hosted inference and from_pretrained loading are not available from this source preview.

Vons studies Direct and Diffusion candidate scoring with abstention. The host owns execution, permission and consent. Model output never authorizes an action. English-first operation and a model asset budget under 64 MiB are design targets; they are not general quality or complete browser-package guarantees.

See the model card, the TypeScript SDK, the paper and Tech Report, and release preparation.

Read, try and contribute

This source preview is distributed through INLEVEL9/Vons. Source and contribution reports belong in inlevel9-com/Vons. The reviewed paper and Tech Report are included as PDF downloads below. The software remains a 0.1.0 research preview. No pretrained weights or working Space are announced.

Follow the community guide for the no-weight walkthrough, experience and reproduction reports, and the contribution guide for review expectations. The paper's arXiv submission record is available. User reports are not independent research validation.

Maintainers can use the Hub upload guide to prepare an organization card, source preview and later a separately validated Space.

Paper and Tech Report

Articles and original figures use CC BY 4.0. Their version, SHA-256 digests and verification scope are in the publication index. These preprints preserve negative results and missing measurements. The paper was submitted to arXiv under submission number 8127259; the last checked status on 2026-09-25 is on hold (undergoing checks). This tracking number is not a public arXiv article identifier. No announcement or peer-review acceptance is claimed.

Local use

python3 -m venv .venv
. .venv/bin/activate
python -m pip install -e '.[dev]'
python -m vons.cli generate-smoke --output data/generated/smoke.jsonl
python -m vons.cli validate-data --input data/generated/smoke.jsonl

The source export includes the research tools and pinned pilot configurations. Obtain and review upstream assets separately before running model experiments. Raw restricted benchmark data and private evidence are not redistributed here.

Terms

Source is distributed under Vons Community and Commercial License 1.0: qualifying noncommercial use is free; enterprise and other commercial use require a separate written agreement. Contact oswarld@inlevel9.com for commercial terms. This is source-available software, not OSI-approved open source.

The paper and Tech Report are distributed under CC BY 4.0. This permits commercial article reuse with attribution; software rights are separate. Third-party assets retain their own terms. See licensing scope. This source release does not establish research validation or a commercial customer agreement.