thox-cluster-swift

THOX configuration of Ministral-3-8B-Instruct-2512 — a THOX configuration and distribution of Ministral 3 8B for local inference and MeshStack deployments: the smallest download in the cluster family (6.0 GB), suited to lighter edge hardware

Live on Ollama

Your AI. Your Data. Your Rules.™

Status: MODEL CARD PAGE — RUNNABLE BUILD ON OLLAMA. This Hugging Face repository is the model's public model card page and THOX packaging record; the runnable build is served from Ollama (see Run it / How to Use). THOX did not train these weights; THOX's work is the configuration, clustering layer, and documentation.

Summary

A THOX configuration and distribution of Ministral 3 8B for local inference and MeshStack deployments: the smallest download in the cluster family (6.0 GB), suited to lighter edge hardware. Configuration-derived: THOX did not train these weights.

Correction notice. This card previously advertised specific tokens-per-second, first-token latency, concurrency, and uptime figures; those were inherited catalog copy with no THOX deployment measurement behind them and have been removed. No benchmark, throughput, latency, concurrency, or uptime figure is claimed for this release. Earlier versions of this card also stated the license as MIT; that was incorrect (see License).

Model ID: Thox-ai/thox-cluster-swift Base model: Ministral 3 8B by Mistral AI Runnable build: Ollama - ollama run Thox-ai/thox-cluster-swift License: Apache License 2.0 (see License)

Highlights

• Configuration, not training: a THOX configuration and distribution of Ministral-3-8B-Instruct-2512 for local inference and MeshStack deployments. THOX did not train these weights; the THOX work is the configuration, clustering layer, and documentation.
• Smallest in the cluster family: 6.0 GB download, suited to lighter edge hardware. Its weight layer is byte-identical to thox-pro-8b's (see What THOX Did and Did Not Do); the two differ in configuration and presentation.
• Runnable build on Ollama: ollama run Thox-ai/thox-cluster-swift. This HF repository is the card page.

Files

MeshStack Origin

This model was designed for MeshStack, THOX's device-to-device clustering technology, built for the flagship ThoxNova devices. ThoxNova hardware was designed to magnetically connect and interact, using custom-developed magnetic pogo pins (MagStack) to distribute models and share compute, memory, and workload from device to device. One device is a computer; a stack of them is a cluster.

That physical stack is the foundation of the MeshStack app and the patent-pending over-air mesh technology THOX ships today, which performs the same device-to-device distribution and resource sharing without a physical connection. MeshStack is included in the THOX Kickstarter campaign (thox.ai).

What THOX Did and Did Not Do

These are THOX-configured model distributions, and the THOX work lives in the MeshStack clustering layer, configuration, documentation, and the origin program, never in the weights. The weight layers in this model's Ollama distribution are byte-identical to the upstream provider's Ollama distribution (SHA-256 verified in the 2026-10-31 release review), and the weights remain governed by the upstream license terms recorded on this card. What THOX built is the layer around them: the MagStack and MeshStack clustering design these configurations were made for, the runtime configuration, and this documentation.

This distribution's weight layer is the same artifact as thox-pro-8b's (identical SHA-256 digest). The two differ in configuration and presentation: Cluster Swift is packaged for multi-device team serving, Pro 8B for single-device use.

Specifications

Intended use

• Interactive local chat and assistance workloads on modest hardware
• High-volume, short-query local inference where a small model is the right tool
• Vision-plus-text tasks such as document and chart reading, within the base model's abilities

Run it

ollama pull Thox-ai/thox-cluster-swift
ollama run Thox-ai/thox-cluster-swift

This Hugging Face repository is the model's public model card page. The runnable build is served from Ollama at https://ollama.com/Thox-ai/thox-cluster-swift - the original card this page is based on lives there in full.

Platforms

Training data and procedure

This is a THOX distribution of the base model named above, packaged and documented by THOX.ai for local and MeshStack/MagStack deployment. The original THOX release materials for this model claim no separate THOX fine-tuning run, training dataset, or training procedure, and none is claimed here. Base-model training data and procedure are those of the base model's provider; see the base model card for that record.

Datasets added by THOX for this release: none.

Evaluation

Benchmark and performance figures quoted in this card were reported by Mistral for the base model and are reproduced from the model's original Ollama model card (catalog-reported at original publication). They were not re-run for this Hugging Face publication.

Independent THOX evaluation of this model on this page: pending. This section will be updated with method and results when a THOX-run evaluation lands; until then, treat quoted figures as the original catalog's claims, not as independently verified measurements.

Catalog notes for this model: No benchmark scores were published on the original card. Performance figures that appeared in earlier catalog copy (tokens-per-second, first-token latency, concurrency, uptime) were removed as unmeasured; THOX claims no throughput, latency, or concurrency figure for this release.

Limitations

• Optimized for speed and rapid queries, not frontier reasoning depth

• Local deployment only in this distribution; behavior and speed depend entirely on the operator's hardware and configuration

• A small model has the reasoning limits of its class; verify important outputs

• All processing is local by design; the model has no access to online information or external services unless the operator wires them in.

• Outputs can be wrong. Professional, medical, legal, financial, and security-relevant outputs require qualified human review.

Hardware Reference (from the original card)

Per device: minimum 16 GB RAM, 15 GB storage, NVIDIA Jetson Orin Nano class GPU (8GB VRAM). Recommended 32 GB RAM, 25 GB NVMe SSD, Jetson Orin AGX class GPU.

Research

THOX MeshStack: Capacity-Priced Pipeline Sharding, Grant-Gated KV Transfer and Lease-Fenced Device Roles for a Private Household Mesh

THOX.ai Research technical report v1.1.0, 4 November 2026. Craig Ross and Tommy Xaypanya, THOX.ai LLC. DOI: https://doi.org/10.5281/zenodo.23969070 — Paper page: https://thox.ai/research/papers/meshstack — PDF: https://thox.ai/papers/meshstack-1.1.0.pdf

License

The base model weights in this distribution are governed by the Apache License 2.0, as recorded in the Ollama manifest license layer for this model. THOX claims no license over the weights beyond the upstream terms. Where a THOX copyright notice appears on a page for this family, it applies to THOX's own material (configuration and documentation) only, not to the weights.

Provenance

• First listed on Ollama under Thox-ai in 2025 and held private.
• Publicly released 2026-10-31 by order of Tommy Xaypanya, CTO and Co-Founder of THOX.ai, as part of the MeshStack model family release.
• Runnable build: Ollama, ollama run Thox-ai/thox-cluster-swift and https://ollama.com/Thox-ai/thox-cluster-swift
• This Hugging Face repository was created 2026-10-31 as the model's public model card page. It contains this card; the runnable model build is served from Ollama.
• Announcement: the MeshStack model family public release announcement (thox.ai).

Citation

@software{thoxai_thox_cluster_swift_2025,
  title = {Thox.ai Cluster Swift - Speed-Optimized MagStack Distributed AI},
  author = {Thox.ai LLC},
  year = {2025},
  url = {https://huggingface.co/Thox-ai/thox-cluster-swift},
  note = {Runnable build: https://ollama.com/Thox-ai/thox-cluster-swift}
}

Your AI. Your Data. Your Rules.™