von-system-one

von-system-one

Docker app from PikkonMG's Repository

Overview

Von System One decision model. A fast local model that answers typed questions about text in one pass: pick a choice, give a score, or say yes or no, each with a calibrated probability. It serves the Jev-compatible /v1/systemone HTTP API for routing, triage, moderation and guardrails.

Before first start:

  1. Pick cpu or nvidia when you install. The nvidia choice sets Extra Parameters to --runtime=nvidia for you and needs the Unraid NVIDIA Driver plugin.
  2. Optional: set an API key. Clients then send Authorization: Bearer YOUR_API_KEY. Leave it blank for no key.
  3. The first start downloads the model weights (about 3 GB) into the Model cache folder. The cpu tag also saves an OpenVINO copy there (about 3 GB more). The container shows healthy when the model is loaded.

The API docs page is at http://YOUR_Unraid_IP:8013/docs. Keep the port on a trusted network.

Von on Unraid

Von is a fast "System One" decision model. You give it some text and a set of typed questions. It answers every question in one pass, each with a calibrated probability:

  • choice picks one option from a list
  • score places the text on a scale
  • noul gives the probability that a statement is true

It serves the Jev-compatible /v1/systemone HTTP API. Use it for routing, triage, moderation and guardrails in front of a larger model.

Von has no official Docker image. This template runs pikkonmg/von-system-one, which is built from the Von release on PyPI and rebuilt when Von has a new release.

First start

  1. Copy templates/von-system-one.xml to /boot/config/plugins/dockerMan/templates-user/my-von-system-one.xml on Unraid. Then open Docker > Add Container and pick von-system-one from the user templates.

  2. Pick a tag. Community Applications asks you when you click Install:

    Tag Extra Parameters Also needed
    cpu Empty Nothing
    nvidia --runtime=nvidia, filled in for you Unraid NVIDIA Driver plugin, driver 560 or newer

    From a user template, set Repository to pikkonmg/von-system-one:nvidia and Extra Parameters to --runtime=nvidia yourself.

    If the nvidia container cannot see the GPU, it stops and the log tells you what is missing.

  3. Optional: set API key. Leave it blank and the API needs no key. When you set one, clients must send it.

  4. Apply the template. The first start downloads the model weights, about 3 GB, into /mnt/user/appdata/von-system-one. The cpu tag also saves an OpenVINO copy there, about 3 GB more. The container shows healthy when the model is loaded.

  5. Open http://YOUR_UNRAID_IP:8013/docs to see the API.

Call the API

Send the text as state and your questions as questions. Leave out the Authorization line when you did not set an API key.

curl -s http://YOUR_UNRAID_IP:8013/v1/systemone \
  -H 'content-type: application/json' \
  -H 'Authorization: Bearer YOUR_API_KEY' \
  --data '{
    "state": {
      "body": "I was charged twice for my subscription. Please refund the duplicate charge."
    },
    "questions": {
      "department": {
        "type": "choice",
        "instructions": "Which department should handle this request?",
        "criteria": {
          "billing": "invoices, payments, refunds",
          "technical": "bugs, outages, system errors",
          "sales": "pricing, new contracts"
        }
      },
      "urgency": {
        "type": "score",
        "instructions": "How urgent is this request?",
        "criteria": ["not urgent", "needs attention soon", "critical deadline or blocking issue"]
      },
      "refund_requested": {
        "type": "noul",
        "instructions": "Does the user explicitly request a refund?"
      }
    }
  }'

GET /health needs no key. It shows the Von version.

Storage and updates

The appdata path maps to /data inside the container. It holds only the downloaded model weights and, on the cpu tag, the OpenVINO copy. You can delete it and both come back on the next start. The container sets the folder owner to PUID:PGID at start.

Set Offline mode to 1 to stop all downloads once the weights are in the cache.

Sources

Install von-system-one on Unraid in a few clicks.

Find von-system-one in Community Apps on your Unraid server, review the template, and click Install. Unraid handles the Docker app or plugin setup from the published template.

Open the Apps tab on your Unraid server Search Community Apps for von-system-one Review the template variables and paths Click Install

Download Statistics

183
Total Downloads

Related apps

Details

Repository
pikkonmg/von-system-one:cpu
Last Updated2026-09-27
First Seen2026-09-27

Runtime arguments

Web UI
http://[IP]:[PORT:8000]/docs
Network
bridge
Shell
bash
Privileged
false

Template configuration

API portPorttcp

Host port for the /v1/systemone API, /health and the /docs page.

Target
8000
Default
8013
Value
8013
Model cachePathrw

Downloaded model weights. The container sets the owner to PUID:PGID at start.

Target
/data
Default
/mnt/user/appdata/von-system-one
Value
/mnt/user/appdata/von-system-one
API keyVariable

Optional. Blank means no key is needed. When set, clients send Authorization: Bearer YOUR_API_KEY. /health never needs a key.

Target
VON_API_KEY
Load model at startVariable

1 loads the model before the server listens, so the first request is fast. 0 loads it on the first request.

Target
VON_PRELOAD
Default
1
Value
1
CPU threadsVariable

Threads for CPU work. Keep at or below your physical core count.

Target
OMP_NUM_THREADS
Default
4
Value
4
Allowed web originsVariable

Web pages allowed to call the API from a browser. * allows all. For a list, separate origins with commas, for example http://192.168.1.10:3000.

Target
VON_CORS_ORIGINS
Default
*
Value
*
Hugging Face tokenVariable

Optional. The model is public and needs no token.

Target
HF_TOKEN
Offline modeVariable

1 uses only the model already in the Model cache and never downloads.

Target
HF_HUB_OFFLINE
Default
0
Value
0
Log levelVariable

One of critical, error, warning, info, debug or trace.

Target
VON_LOG_LEVEL
Default
info
Value
info
NVIDIA GPUVariable

nvidia tag only. Use all or one GPU UUID from the NVIDIA Driver plugin page.

Target
NVIDIA_VISIBLE_DEVICES
Default
all
Value
all
NVIDIA capabilitiesVariable

nvidia tag only. Keep compute,utility.

Target
NVIDIA_DRIVER_CAPABILITIES
Default
compute,utility
Value
compute,utility
PUIDVariable

User the server runs as. Unraid default is 99 (nobody).

Default
99
Value
99
PGIDVariable

Group the server runs as. Unraid default is 100 (users).

Default
100
Value
100