BitNet s9pk Package

StartOS package for BitNet LLM with SSH access.

Project Structure

bitnet-s9pk/
├── Dockerfile              # Extended BitNet image with SSH
├── docker_entrypoint.sh    # Entrypoint script for SSH setup
├── manifest.yaml           # StartOS package manifest
├── instructions.md         # User-facing instructions
├── Makefile               # Build automation
├── scripts/
│   ├── config_get.sh      # Config schema
│   ├── config_set.sh      # Config application
│   ├── properties.sh      # Package properties
│   └── health_check.sh    # Health check script
└── assets/
    └── icon.png           # Package icon (TODO)

Prerequisites

  • Docker access (local or remote via SSH to MacBook)
  • start-cli installed and authenticated
  • Developer key at /data/.startos/developer.key.pem

Building

cd /data/.openclaw/workspace/bitnet-s9pk
make

This will:

  1. Build the Docker image on remote Docker (MacBook)
  2. Export it as image.tar
  3. Pack everything into bitnet.s9pk

Option 2: Manual Build

# Build image on MacBook
docker -H ssh://macbook build --platform linux/amd64 \
  --tag start9/bitnet/main:1.0.0 \
  -o type=docker,dest=image.tar .

# Pack the s9pk
start-cli s9pk pack --arch aarch64 --arch x86_64 .

Installing

# Install on your StartOS server
start-cli package install bitnet.s9pk

# Or use make
make install

Configuration After Install

  1. Go to BitNet service in StartOS UI
  2. Navigate to Config tab
  3. Paste your SSH public key
  4. Save and restart the service

Connecting via SSH

# Via Tor (get address from Interfaces tab)
ssh -o ProxyCommand="nc -x localhost:9050 %h %p" root@your-bitnet-address.onion

# Via LAN
ssh root@your-server-lan-ip -p <port>

Using BitNet

Once connected:

# Run inference
python3 /BitNet/run_inference.py -m ggml-model-i2_s.gguf -p "Hello, world!"

# See help
python3 /BitNet/run_inference.py --help

TODO

  • Add proper icon.png (512x512)
  • Test on actual StartOS hardware
  • Add model download action
  • Volume mount for custom models

References

Description
No description provided
Readme 29 KiB
Languages
Shell 49.9%
Dockerfile 26.8%
Makefile 23.3%