Model Catalog
Published THOX models on Hugging Face and Ollama. Pro 8B-32B cards are unpublished and removed from this page.
Published now
Pull from the public orgs. These ids were checked on 2026-09-29.
- ollama.com/Thox-ai — thox-ai/thoxforge-7b, thoxwave-8b, thoxglobal-7b, thoxmini-3b, thoxnova-12b-core, thoxnova-12b-agent, thoxroute:1.5b, plus device-role tags.
- huggingface.co/Thox-ai — 39 public models including ThoxKey-9M-role, ThoxAir-16M-role, ThoxClip-9M-role, ThoxMini-125M-role, ThoxLLM-327M-v2, ThoxRoute-1.5B.
There is no public thox-ai/thoxforge-7b through thox-ai/thoxwave-8b repository. ThoxNova in those orgs is a model family, not a campaign device.
The Pro cards that follow are a planned professional lineup. They are not current Ollama or Hugging Face install names. Local-mode processing, when a model is actually installed, stays on hardware you control.
Healthcare
Records stay on local hardware
Legal
Attorney-Client Privilege
Enterprise
No data leaves the operator’s control
Research
FERPA-Aware Controls
Recommended for Most Users
Start with public pulls such as thox-ai/thoxforge-7band the device role cards on Hugging Face. THOX Pro 8B-32B is unpublished and is not listed here.
Published models
ThoxForge 7B
RecommendedPublishedthox-ai/thoxforge-7b
Public coding and reasoning assistant. Mistral-7B-v0.3 QLoRA, Q4_K_M on Ollama.
Size
7B
Memory
See Ollama tag
Speed
Not published
Context
See model card
Base Model
Mistral-7B-Instruct-v0.3
Key Industries:
Key Capabilities:
- Public on Ollama and Hugging Face
- Host-class coding assistant
- Not a campaign device SKU
ThoxMini 3B
Publishedthox-ai/thoxmini-3b
Public host-class Mini assistant. Distinct from ThoxMini-125M-role on-device card.
Size
3B
Memory
See Ollama tag
Speed
Not published
Context
See model card
Base Model
Llama-3.2-3B-Instruct
Key Industries:
Key Capabilities:
- Public Ollama pull
- Companion for Mini hosts
- Not the on-device 125M role card
ThoxRoute 1.5B
Publishedthox-ai/thoxroute
Public routing classifier. Clean-room Qwen2.5-1.5B fine-tune.
Size
1.5B
Memory
~1 GB Q4
Speed
Not published
Context
32K on the Ollama tag
Base Model
Qwen2.5-1.5B-Instruct
Key Industries:
Key Capabilities:
- Public router
- Pairs with device role cards
- Infrastructure, not a chat SKU
ThoxKey 9M role
Device roleThox-ai/ThoxKey-9M-role
Public Key role card on Hugging Face. Host still required for 7B-class work.
Size
9M
Memory
On-device role
Speed
Not published
Context
See model card
Base Model
ThoxMicro-1bit-9M
Key Industries:
Key Capabilities:
- Public HF weights
- Mapped to ThoxKey
- Capability cert is public+mapped only
ThoxAir 16M role
Device roleThox-ai/ThoxAir-16M-role
Public Air role card. Private thox-gem-e4b is not a public Air install.
Size
16M
Memory
On-device role
Speed
Not published
Context
See model card
Base Model
Thox role card
Key Industries:
Key Capabilities:
- Public HF weights
- Mapped to ThoxAir
- No unpublished E4B claim
Installing Models
# Public Ollama library
$ ollama pull thox-ai/thoxforge-7b:Q4_K_M
$ ollama pull thox-ai/thoxmini-3b:Q4_K_M
$ ollama pull thox-ai/thoxroute:1.5b
# Device role tags (also on Hugging Face)
$ ollama pull thox-ai/thoxair-16m-role
$ ollama pull thox-ai/thoxclip-9m-role
# Check installed models
$ ollama list
Quantization Options
Quality vs. Performance Tradeoffs
Each model is available in different quantization levels, trading off quality for speed and memory usage.
| Quantization | Quality | Speed | Memory (7B) |
|---|---|---|---|
| fp16 | Highest | Slowest | ~14GB |
| q8_0 | Very High | Fast | ~7GB |
| q4_k_m | Good | Very Fast | ~4GB |
| q4_0 | Acceptable | Fastest | ~3.5GB |
Recommendation: For most users, q4_k_m offers the best balance of quality and performance. Use q8_0 if you need higher quality and have memory to spare.
Model Selection Guide
| Use Case | Recommended | Memory | Why |
|---|---|---|---|
| Everyday professional AI | thox-ai/thoxforge-7b | 10GB | Public host assistant. Speed not certified. |
| Complex visual analysis | Thox-ai/ThoxAir-16M-role | 14GB | Advanced vision + reasoning |
| Software engineering | thox-ai/thoxforge-7b | 20GB | Agentic coding, tool use, 384K context |
| Advanced research & math | thox-ai/thoxnova-12b-core | 24GB | Host-class family model. No published AIME/GPQA certificate. |
| Document OCR (32 languages) | thox-ai/thoxwave-8b | 24GB | Vision specialist, scanned documents |
Local Privacy Boundaries
In local mode, THOX Pro inference runs on your THOX.ai device without sending prompts or model outputs to a remote inference provider. Framework tags below are deployment alignment targets, not certifications; each organization must validate its complete system.
CONFIDENTIAL AND PROPRIETARY INFORMATION
This documentation is provided for informational and operational purposes only. The specifications and technical details herein are subject to change without notice. THOX.ai LLC reserves all rights in the technologies, methods, and implementations described.
Nothing in this documentation shall be construed as granting any license or right to use any patent, trademark, trade secret, or other intellectual property right of THOX.ai LLC, except as expressly provided in a written agreement.
Patent Protection
The MagStack™ magnetic stacking interface technology is proprietary technology of THOX.ai LLC, protected by trade secrets and intellectual property laws....
Reverse Engineering Prohibited
Except as permitted by applicable law or a component’s applicable license, you may not reverse engineer, disassemble, decompile, decode, or otherwise attempt to derive the source c...
THOX.ai™, ThoxOS™, MagStack™, MeshStack™, ThoxMigrate™, the THOX Edge Series™, the THOX Nova Series™, and the THOX.ai logo are trademarks of THOX.ai LLC. WireGuard® is a registered trademark of Jason A. Donenfeld.
All other trademarks are the property of their respective owners.
© 2026 THOX.ai LLC. All Rights Reserved.