Instructions to use PINKlab/OSHE-Local-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use PINKlab/OSHE-Local-v1 with MLX:
# Make sure mlx-lm is installed # pip install --upgrade mlx-lm # if on a CUDA device, also pip install mlx[cuda] # Generate text with mlx-lm from mlx_lm import load, generate model, tokenizer = load("PINKlab/OSHE-Local-v1") prompt = "Once upon a time in" text = generate(model, tokenizer, prompt=prompt, verbose=True) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- MLX LM
How to use PINKlab/OSHE-Local-v1 with MLX LM:
Generate or start a chat session
# Install MLX LM uv tool install mlx-lm # Generate some text mlx_lm.generate --model "PINKlab/OSHE-Local-v1" --prompt "Once upon a time"
- Atomic Chat
OSHE Local v1 Community Preview
OSHE Local v1 is a local AI engine for Apple Silicon Macs, published by PINKlab / byPINKLAND Limited. It exposes an OpenAI-compatible loopback API and includes the public OSHE multi-document capability pack selected by the runtime when that task requires it.
This repository contains the free Community Preview v0.10 binary package. It is an engine-only release: Tabi is not required, and Tabi, Pet n Kiss, private product connectors, training data, credentials, and paid services are not included.
Release facts
| Item | Value |
|---|---|
| Release | Community Preview v0.10 (0.1.0-preview.6) |
| Platform | macOS on Apple Silicon (arm64) |
| API base | http://127.0.0.1:8093/v1 |
| Model ID | oshe-local-v1 |
| Archive size | 2,534,420,369 bytes |
| Archive SHA-256 | 757bc8542a0ccfb9cf2242ce961b727ee1b7255aee7921a17a28cb4db7813822 |
| Publisher | byPINKLAND Limited, company no. 13945554 |
The release envelope is signed with Ed25519. The included installer verifies the frozen bundle inventory and all required capability packs before activation.
What is included
- Gemma 4 E2B foundation lineage packaged for local MLX inference.
- Local chat through
/v1/chat/completions. - OSHE host capabilities for browser-action policy, failure-repair policy, filename search, memory, search synthesis, and source-grounded multi-document handling.
- One public, signed learned capability pack:
oshe-owner-multidoc-public-capability-pack-v0version1.0.0. - A self-contained CPython runtime and frozen dependency inventory for macOS arm64.
OSHE selects zero or one learned capability pack per task. Users do not install adapters separately.
Install
Download OSHE-Local-v1-Community-Preview-v0-10-signed-rc.tar.gz, verify its SHA-256, then follow INSTALL.md. The extracted package also includes a standalone signature verifier with a pinned trusted-key fingerprint.
After installation:
curl http://127.0.0.1:8093/health
curl http://127.0.0.1:8093/v1/models
OpenAI-compatible chat example:
curl http://127.0.0.1:8093/v1/chat/completions \
-H 'Content-Type: application/json' \
-d '{
"model": "oshe-local-v1",
"messages": [{"role": "user", "content": "用繁體中文簡單介紹你自己。"}]
}'
Preview status
This is a zero-price Community Preview for evaluation, testing, feedback, and building company references. It is not a support, uptime, performance, security, or fitness-for-purpose commitment. The service binds to loopback only by default.
Before installing, read:
The OSHE-owned portions are not released under an open-source licence. Third-party components remain governed by their own licences and notices inside the package.
Contact
Product and preview enquiries: hello@bypinkland.com