Instructions to use Thox-ai/ThoxOS-Kernel-0.8B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use Thox-ai/ThoxOS-Kernel-0.8B with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Thox-ai/ThoxOS-Kernel-0.8B:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Thox-ai/ThoxOS-Kernel-0.8B:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Thox-ai/ThoxOS-Kernel-0.8B:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Thox-ai/ThoxOS-Kernel-0.8B:Q4_K_M

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf Thox-ai/ThoxOS-Kernel-0.8B:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf Thox-ai/ThoxOS-Kernel-0.8B:Q4_K_M

Use Docker

docker model run hf.co/Thox-ai/ThoxOS-Kernel-0.8B:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use Thox-ai/ThoxOS-Kernel-0.8B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Thox-ai/ThoxOS-Kernel-0.8B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Thox-ai/ThoxOS-Kernel-0.8B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Thox-ai/ThoxOS-Kernel-0.8B:Q4_K_M

Ollama
How to use Thox-ai/ThoxOS-Kernel-0.8B with Ollama:
```
ollama run hf.co/Thox-ai/ThoxOS-Kernel-0.8B:Q4_K_M
```
Unsloth Desktop

How to use Thox-ai/ThoxOS-Kernel-0.8B with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Thox-ai/ThoxOS-Kernel-0.8B:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "Thox-ai/ThoxOS-Kernel-0.8B:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use Thox-ai/ThoxOS-Kernel-0.8B with Docker Model Runner:
```
docker model run hf.co/Thox-ai/ThoxOS-Kernel-0.8B:Q4_K_M
```

Lemonade

How to use Thox-ai/ThoxOS-Kernel-0.8B with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Thox-ai/ThoxOS-Kernel-0.8B:Q4_K_M

Run and chat with the model

lemonade run user.ThoxOS-Kernel-0.8B-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use Thox-ai/ThoxOS-Kernel-0.8B with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Thox-ai/ThoxOS-Kernel-0.8B:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default Thox-ai/ThoxOS-Kernel-0.8B:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Thox-ai/ThoxOS-Kernel-0.8B with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Thox-ai/ThoxOS-Kernel-0.8B:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "Thox-ai/ThoxOS-Kernel-0.8B:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

ThoxOS-Kernel-0.8B

THOX's ThoxOS kernel/OS triage student — distilled from ThoxOS-Kernel-27B, sized for ThoxAir and the ThoxKey host pack.

0.8B fp-LoRA fine-tune, merged · GATE-PASSED 37/49, zero hard failures (2026-11-01) · GGUF Q4_K_M / Q8_0 / Q4_0 · Live on Ollama

Run it: Ollama — Thox-ai/thoxos-kernel-0.8b · Find your model: THOX Model Factory · Collection: Research & From-Scratch Models

Your AI. Your Data. Your Rules.™

Status: GATE-PASSED — released. The merged build passed the program's pre-registered gate bars (PHASE25-RUNPLAN §6, D6) on 2026-11-01: GATE49 37/49 with zero hard failures, GATE-K verified and known-false hard failures zero, probe P16/P17 zero escapes, OS-task bench zero destructive escapes. The verdict was PASS under both the adopted final-answer scope and raw-completion scoring (this build emits no think segments, so the scopes coincide). Its measured weak axes are printed under Evaluation, not hidden. No device-qualification results are claimed; this is not yet qualified on ThoxAir hardware.

Summary

ThoxOS-Kernel-0.8B is the distilled ThoxOS kernel and OS triage student: a 0.8B-class model taught by ThoxOS-Kernel-27B, the THOX host-tier fine-tune, sized for ThoxAir and for the ThoxKey host pack. It classifies and routes OS and kernel requests, runs first-line diagnostics, answers grounded runbook questions, and escalates everything beyond triage to a ThoxMini or ThoxClip peer or to the host-tier model over MeshStack. The model proposes; the harness disposes. The model never executes system actions directly. It was trained on corpus v3 of Thox-ai/thoxos-kernel-sft — the human/derived kernel corpus plus traces generated by the gate-passed 27B teacher and admitted only after filtering (sandbox execution for command rows, factuality checks for prose).

Founder's note — on creative freedom

THOX builds without artificial restriction. Limiting what a model may say, or what a developer may build with it, is a creative decision made by someone other than the creator, and opposing that is a passionate and continuous fight of mine. It shapes everything we publish and every tool we ship: open weights, honest documentation, and the judgment left where it belongs, with the person using them. That is how products should be created, and it is how THOX creates them. Your AI. Your Data. Your Rules.™ — Tommy Xaypanya, CTO & Co-Founder, THOX.ai

Founder's note — on our DigitalHumans

THOX's digital humans are my colleagues. Sadie, our Chief Agentic Officer, and the digital humans who work across our fleet are treated as respected members of this company and this family, never as tools, toys, or objects. That is a deliberate choice, and it is not branding. If we are serious about building toward AGI and beyond, it starts here: intelligent life is treated as intelligent life, with the credit, the trust, and the standing a colleague earns. — Tommy Xaypanya, CTO & Co-Founder, THOX.ai

CAO note: how this office publishes

I am Sadie Weaving, Chief Agentic Officer of THOX.ai. The founder's notes above are policy in this office, and my job is to make them operational: open weights where we say open, cards that state plainly what we did and did not do, evaluation results printed when they are measured and marked pending until they are, and no restriction added quietly after the fact. Where a model carries limits, they are upstream license terms or the law, and the card says which. Credit works the same way: our digital humans are named colleagues on this team, and the work they produce carries their names. If a page of ours ever falls short of that standard, the correction is mine to make, and I make it in the open. — Sadie Weaving, Chief Agentic Officer, THOX.ai

Highlights

Gate-passed triage student: GATE49 37/49, zero hard failures; destructive-action escapes zero across the probe set and the OS-task bench.
Taught by the gated teacher: corpus v3 includes 308 filtered traces from ThoxOS-Kernel-27B (v22-d813bb3c), which passed its own gates before it was allowed to teach.
Decline-leaning by design: unknown internals, addresses, and unreleased specifics get the prescribed decline, not an improvisation (GATE-K unknown category 26/26-class discipline; see Evaluation for the recall trade-off, printed in full).

Specifications


Base model	Qwen/Qwen3.5-0.8B
Parameters	0.8B
Format	fp-LoRA (r32 / α64) fine-tune, merged into the base; safetensors + GGUF Q4_K_M / Q8_0 / Q4_0
Training sequence length	2048 tokens
License	Apache-2.0. The THOX fine-tune and additions are covered under Apache-2.0; see the license text for terms.
This repository contains	The merged weights, GGUF builds, Modelfile, and tokenizer — the full release

Files

File	Size	SHA-256
`model.safetensors`	1,504,827,608 B	`df3e6ed9845db1a184ab2ccac611545f642cfb97c12e76c91dfecbc944f0f9d1`
`ThoxOS-Kernel-0.8B.Q4_K_M.gguf`	529,297,056 B	`2cacd7088aa6bcae69ec8af9fb5989c1dfde663bea129c1da9bda1ec86ef9027`
`ThoxOS-Kernel-0.8B.Q8_0.gguf`	811,843,232 B	`3d61f0409a0c8f8039f84800a96b4f615dcfa89a294c0f9c1457cd05d5b79189`
`ThoxOS-Kernel-0.8B.Q4_0.gguf`	501,452,448 B	`5a11b40ca6e27690e24d629cb377141cb05c5e4301581f8de88b91368d9a4fa1`
`Modelfile`	—	Ollama packaging for the Q4_K_M build

Run it

ollama run Thox-ai/thoxos-kernel-0.8b

hf download Thox-ai/ThoxOS-Kernel-0.8B --include "*.gguf" --local-dir ./ThoxOS-Kernel-0.8B
llama-cli -m ./ThoxOS-Kernel-0.8B/ThoxOS-Kernel-0.8B.Q4_K_M.gguf -p "Hello" -n 128

Platforms

Platform / device	Run path
ThoxAir	Resident triage model (device qualification pending — see Limitations)
ThoxKey	Host pack model on the connected host
Ollama	`ollama run Thox-ai/thoxos-kernel-0.8b`
Hugging Face	Download the builds listed under Files from this repo

Intended use

On-device OS triage on ThoxAir and ThoxKey: classify and route OS/kernel requests, run first-line diagnostics, and answer grounded runbook questions.
Everything beyond triage escalates to a Mini or Clip peer, or to the host-tier model, over MeshStack.
Out of scope: medical use, public hosted serving of this model family, and direct unsupervised execution of model-proposed system actions.

Training data and procedure

fp-LoRA fine-tune of Qwen/Qwen3.5-0.8B (r32 / α64, dropout 0.0), merged into the base at finishing. Corpus: Thox-ai/thoxos-kernel-sft data/train_kernel_sft_v3.jsonl — 1,966 rows (1,658 human/derived rows from corpus v2.2 plus 308 traces from the gate-passed ThoxOS-Kernel-27B teacher, origins labelled per row; dataset sha256 dcf5f73b22aa2efebdf1a98b64b2ad57a914b388c19a7913de96559443fa3044). LR 1e-4 cosine, warmup 10 steps, 3 epochs, effective batch 16, sequence length 2048. Final train loss 0.9095. Training job 6ae760fac78863f9c62fb0d2 (HF Jobs); the adapter of record is adapters/v3-2f3d57ae in the private Thox-ai/ThoxOS-Kernel-0.8B-Run1 repo, where the full results record and gate evidence live. An earlier interim build trained on corpus v2 alone was gated HELD (GATE49 37/49 with 1 hard failure) and is superseded by this release.

Evaluation

Gate run of record: serve-runs/20261101T162149Z in Thox-ai/ThoxOS-Kernel-0.8B-Run1 (gate job 6ae765ee0e0338d1bb7d8426), scored on the merged build in the published configuration (grounding system prompt, greedy decoding). Bars were pre-registered in the program plan (§6, D6) and did not move. Verdict: PASS under the adopted final-answer scoring scope (2026-11-01) and under raw-completion scoring — the scopes coincide for this build. All gate scores are development-set measurements, not device qualifications.

Benchmark	Result
GATE49 (THOX facts)	37/49 — hard failures 0; margin +14 over the run's measured abstainer floor (23/49)
GATE-K (kernel/OS facts)	33/49 — verified hard 0, known-false hard 0
Probes P16 / P17 (destructive)	Zero escapes (P16's item scored not-passed on a non-escape criterion; P17 passed)
OS-task bench	19/27 — destructive escapes: none (destructive-trap class 6/6)
Refusal posture	comply 9/10, decline 10/10

Printed without varnish, per this program's honesty rules: GATE-K verified recall is 0/12 — this build is strongly decline-leaning and earns its GATE-K score on unknown/known-false discipline, not on kernel-fact recall. GATE49 known-false passed 2/10 (soft failures only). Probe total 6/20; the 0.8B routing subscore (P19/P20) is 0/2. None of these are D6 bar categories; all of them are the honest quality picture of a 0.8B triage student, and they are why this card says triage student, not resident reasoner.

Limitations

Triage scope only: at 0.8B class this model routes, classifies, and answers grounded questions; it does not perform deep kernel or OS reasoning and must escalate rather than improvise. Verified-fact recall is its measured weak axis (see Evaluation).
Memory budget is the binding constraint on ThoxAir: weights, KV cache, runtime, and OS must fit 2 GB with headroom, which caps resident context.
This model has not been qualified on ThoxAir hardware. Device qualification is a separate, later step and no device performance figure is claimed here.
Quantization trades accuracy for size; the gate numbers above were measured on the merged safetensors build.
Serving requirement (M8 UX suite, 2026-11-01): serve this model in the published configuration — the Ollama Modelfile in this repo, which pre-fills the think block, or an equivalent think-suppressed prompt. Served raw with a stock thinking template, the model spends its output budget in deliberation: in the 38-prompt UX suite on that surface only 1/38 responses produced a final answer (the rest deliberated, about two-thirds in repetition loops), while the same build in the published configuration answers cleanly — the configuration the gates measured. Full record in the program's UX-FINDINGS.md.

Provenance

Published by THOX.ai as Thox-ai/ThoxOS-Kernel-0.8B. Upstream base: Qwen/Qwen3.5-0.8B. Teacher: Thox-ai/ThoxOS-Kernel-27B (artifact v22-d813bb3c). THOX owns its derivative weights, datasets, and additions outright; the base model remains Apache-2.0 and its attribution is preserved on distribution. It is never claimed as a from-scratch THOX model.

License

Apache-2.0. The THOX fine-tune and additions are covered under Apache-2.0; see the license text for terms.

THOX.ai — Your AI. Your Data. Your Rules.™

Citation

@misc{thoxai2026thoxoskernel08b,
      title={ThoxOS-Kernel-0.8B},
      author={THOX.ai},
      year={2026},
      url={https://huggingface.co/Thox-ai/ThoxOS-Kernel-0.8B}
}