# Modelship CPU

Modelship CPU is a self-hosted, OpenAI-compatible AI inference server that serves multiple reasoning LLM, embedding, speech, and image models on CPU behind one gateway. It is open source and self-hosted, installable on Unraid.

Canonical page: https://ourlocalhost.com/app/3064/modelship-cpu

## Description

Modelship CPU is a self-hosted, OpenAI-compatible AI inference server that serves multiple reasoning LLM, embedding, speech, and image models on CPU behind one gateway.

## Platform availability

Available on 1 of 10 platforms: Unraid.

## Health score

56/100 (Fair), based on 4 of 5 factors.
- Maintained: 100 (last commit 6 days ago)
- Popular: 35 (55 GitHub stars)
- Easy to install: 35 (1 install platform, install notes, Docker image)
- Security check: 15 (2 serious findings, 9 minor, CI, tests)

## Security check

2 serious findings: Uses anyio 4.13.0, which has a known critical vulnerability; Uses certifi 2025.11.12, which has a known critical vulnerability.
Automated checks on commit 44334a2, 13 Nov 2026. Not an audit.

## Links

- Unraid: https://ca.unraid.net/apps/modelship-cpu-1bf5ye11uqpdot?q=modelship-cpu
- Official website: https://docs.model-ship.ai/
- GitHub: https://github.com/modelship-ai/modelship

## Common questions

### Can I run a single model without creating models.yaml?

The quick start deploys one model with the --model flag. Use models.yaml when deploying several models at once.

### How do I persist downloaded model files?

Mount a named Docker volume to /.cache, such as modelship-cache, so the model cache survives container recreation.

### Which port and API base path should clients use?

The CPU example maps host port 8000 to container port 8000. Requests target the /modelship/v1 base path, including /modelship/v1/responses and /modelship/v1/chat/completions.

### How many CPU cores should I reserve for a model?

The example reserves four cores with --num-cpus 4. Lower the value if the container has fewer cores because the deploy waits for resources it cannot get.

### Does the CPU variant require GPU software?

The CPU quick start uses the latest-cpu image and does not pass GPU flags. The README says NVIDIA GPU deployment needs the NVIDIA Container Toolkit.

---
Source: Our Local Host (https://ourlocalhost.com/app/3064/modelship-cpu), the self-hosted and open-source app directory.
