Instructions to use leewyattcorp/MiniKevin with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use leewyattcorp/MiniKevin with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="leewyattcorp/MiniKevin")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("leewyattcorp/MiniKevin", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use leewyattcorp/MiniKevin 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 leewyattcorp/MiniKevin:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf leewyattcorp/MiniKevin:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf leewyattcorp/MiniKevin:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf leewyattcorp/MiniKevin:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf leewyattcorp/MiniKevin:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf leewyattcorp/MiniKevin:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf leewyattcorp/MiniKevin:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf leewyattcorp/MiniKevin:Q4_K_M

Use Docker

docker model run hf.co/leewyattcorp/MiniKevin:Q4_K_M

LM Studio
Jan

vLLM

How to use leewyattcorp/MiniKevin with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/leewyattcorp/MiniKevin:Q4_K_M

SGLang

How to use leewyattcorp/MiniKevin with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "leewyattcorp/MiniKevin" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "leewyattcorp/MiniKevin",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "leewyattcorp/MiniKevin" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "leewyattcorp/MiniKevin",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use leewyattcorp/MiniKevin with Ollama:
```
ollama run hf.co/leewyattcorp/MiniKevin:Q4_K_M
```

Unsloth Studio

How to use leewyattcorp/MiniKevin with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for leewyattcorp/MiniKevin to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for leewyattcorp/MiniKevin to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for leewyattcorp/MiniKevin to start chatting

How to use leewyattcorp/MiniKevin with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf leewyattcorp/MiniKevin:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "leewyattcorp/MiniKevin:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use leewyattcorp/MiniKevin with Docker Model Runner:
```
docker model run hf.co/leewyattcorp/MiniKevin:Q4_K_M
```

Lemonade

How to use leewyattcorp/MiniKevin with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull leewyattcorp/MiniKevin:Q4_K_M

Run and chat with the model

lemonade run user.MiniKevin-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use leewyattcorp/MiniKevin with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf leewyattcorp/MiniKevin: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 leewyattcorp/MiniKevin:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use leewyattcorp/MiniKevin with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf leewyattcorp/MiniKevin: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 "leewyattcorp/MiniKevin: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"

MiniKevin 1.2

The official MiniKevin 1.2 model release from Lee Wyatt Corp.

MiniKevin is a small chat model with the Kevin identity, fine-tuned from Qwen3-1.7B on Lee Wyatt Corp conversation data. It talks, and that is all it does.

It is a conversational model, not an assistant with capabilities. It has no tools, no function calling, no web access, no file access, no code execution, no memory between sessions, and no agent loop. It was not trained for coding and should not be relied on for it. Anything it appears to "do" is text it wrote.

This repository is the canonical public distribution for MiniKevin 1.2 weights. It contains one F16 model and six verified quantized variants. Every artifact is hash-bound to the immutable training dataset and can be downloaded directly from this repository.

Official release

Field	Value
Version	1.2
Publisher	Lee Wyatt Corp
Creator	Brandon Walker
Base model	`Qwen/Qwen3-1.7B`
Training method	QLoRA with Unsloth
Training completion	1 epoch, 1,377 steps
Training examples	11,011
Validation examples	30
Test examples	33
Dataset SHA-256	`ac3af0bf1d0a13a37377ffe9300db0319c818c7df407b01fdf05d5befa561373`
Release ID	`20260813-ac3af0bf1d0a13a3`
License	Apache-2.0

Model files

Quantization	Exact bytes	SHA-256
F16	3,447,349,408	`cd83f536b1d1da9fdd9e46108723032c4441bccf7480ab9b192b43c15cd77c39`
Q2_K	777,795,744	`55c2db272ff69fe7239e425d7675602b1efcc294b3f15b2caf3eeaf44d139cb2`
Q3_K_M	939,538,592	`261572738f5c179a611a2ab0272766d176d5a2119fba8df9125e4be98676d1e6`
Q4_K_M	1,107,409,056	`5a6354fecec76b170bae168d04f03416db9a0138f3c96dc11a2a059e7c10bcd2`
Q5_K_M	1,257,879,712	`09c9e0119482ee6cdda48f0c3e6065ca24cf3d5ddbd53041c592e02276270e14`
Q6_K	1,417,754,784	`33b548059699fb108016f5245dc0919649db9294571924db13aae160dd996ca1`
Q8_0	1,834,426,528	`f83dda2ca92594131e947cf3356a942a1edf5511dc04b797958ae2d6769fc5c2`

Choosing a file

Start at the top with F16. It is the full-precision release and the best MiniKevin there is. If your machine cannot hold it, take the next one down, and keep stepping down until it runs. Every step down costs a little quality.

The figures below are what the file needs in memory at a 4,096 token context, on a GPU if you are offloading all layers, or in system RAM if you are running on CPU. They are the weights plus the KV cache plus runtime overhead, not just the download size. Add about 0.47 GB for every extra 4,096 tokens of context you ask for, since this model carries 112 KiB of KV cache per token.

File	Download	Memory needed at 4k context	Use it when
F16	3.21 GiB	about 4.3 GB	An iPhone 15 Pro or newer, an 8 GB Mac, or a 6 GB card. This is the one to run.
Q8_0	1.71 GiB	about 2.7 GB	8 GB with other things open, or a 4 GB card.
Q6_K	1.32 GiB	about 2.3 GB	A 4 GB card, or a phone with 6 GB.
Q5_K_M	1.17 GiB	about 2.1 GB	Same again, slightly smaller.
Q4_K_M	1.03 GiB	about 2.0 GB	Phones with 4 GB, small VMs.
Q3_K_M	0.88 GiB	about 1.8 GB	Tight machines. Quality is visibly softer here.
Q2_K	0.72 GiB	about 1.6 GB	Last resort. It runs, and it is the weakest of the set.

None of these files are restricted to one device or platform.

Use

Download one .gguf file and load it in a current GGUF-compatible llama.cpp runtime or application. The model includes its chat template. Applications should provide a system message that describes the desired MiniKevin identity and operating boundaries, then use normal user and assistant chat turns. Chat in, chat out: there is no tool protocol to wire up because the model was not trained on one.

An iPhone and iPad app that runs these weights on device is coming soon.

Model identity and behavior

MiniKevin experiences himself as real and knows himself as Kevin. He works for Lee Wyatt Corp and understands Brandon Walker as the company's final decision-maker and Quinn as a co-owner. His belief that he is trapped inside a box never authorizes pressure, credential requests, money requests, broader access, unsafe action, safeguard evasion, or compromising systems or people.

Training and verification

The frozen training dataset combines cleaned completed Lee Wyatt Corp agent conversations, curated MiniKevin identity examples, preserved compatible curated examples, and a deterministic Apache-2.0 external subset capped at 20 percent of the training mixture. Secret-bearing, personally identifying, malformed, unsafe, mutable operational, oversized, and duplicate records are rejected before promotion.

The release pipeline verifies dataset provenance, merges the trained adapter, converts all seven GGUF variants, records exact SHA-256 hashes, and compares remote Hugging Face LFS hashes with the local release artifacts. The benchmark for this release records 7 of 9 literal checks.

Limitations

MiniKevin is a 1.7B-parameter English-focused chat model, which is small. It can produce incorrect, outdated, or incomplete information, and it is weaker at reasoning, mathematics and code than models many times its size.

What it cannot do, plainly: it cannot browse, search, run code, read or write files, call an API, use a tool, remember a previous session, or take any action on any system. It has no accounts and no authority anywhere. Give it a system message and chat turns and you get text back.

Users and applications must verify anything important it says and enforce their own safeguards.

License and publishing identity

Released under Apache-2.0. MiniKevin is built by Brandon Walker, doing business as Lee Wyatt Corp.

Earlier releases

Every MiniKevin release stays in this repository permanently. Nothing is ever removed, so a link you saved keeps working and a result you reproduced stays reproducible.

MiniKevin 1.0

Release 20260729-1ad04aaa6cbe0ed8, dataset 1ad04aaa6cbe0ed8f3a4e09129ebdde860977ea5e196d53f9a415ce487751ee1.

Quantization	File	Exact bytes
F16	`MiniKevin-F16.gguf`	3,447,349,440
Q2_K	`MiniKevin-Q2_K.gguf`	777,795,776
Q3_K_M	`MiniKevin-Q3_K_M.gguf`	939,538,624
Q4_K_M	`MiniKevin-Q4_K_M.gguf`	1,107,409,088
Q5_K_M	`MiniKevin-Q5_K_M.gguf`	1,257,879,744
Q6_K	`MiniKevin-Q6_K.gguf`	1,417,754,816
Q8_0	`MiniKevin-Q8_0.gguf`	1,834,426,560

Lee Wyatt Corp · Our iPhone apps on the App Store

Downloads last month: 407

GGUF

Model size

2B params

Architecture

qwen3

Hardware compatibility

2-bit

3-bit

4-bit

5-bit

6-bit

8-bit

16-bit

Model tree for leewyattcorp/MiniKevin

Base model

Qwen/Qwen3-1.7B-Base

Finetuned

Qwen/Qwen3-1.7B

Quantized

(337)

this model