Instructions to use KOHENOOR-AI/kai-lite2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use KOHENOOR-AI/kai-lite2 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 KOHENOOR-AI/kai-lite2:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf KOHENOOR-AI/kai-lite2:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf KOHENOOR-AI/kai-lite2:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf KOHENOOR-AI/kai-lite2: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 KOHENOOR-AI/kai-lite2:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf KOHENOOR-AI/kai-lite2: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 KOHENOOR-AI/kai-lite2:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf KOHENOOR-AI/kai-lite2:Q4_K_M

Use Docker

docker model run hf.co/KOHENOOR-AI/kai-lite2:Q4_K_M

LM Studio
Jan

vLLM

How to use KOHENOOR-AI/kai-lite2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "KOHENOOR-AI/kai-lite2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "KOHENOOR-AI/kai-lite2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/KOHENOOR-AI/kai-lite2:Q4_K_M

Ollama
How to use KOHENOOR-AI/kai-lite2 with Ollama:
```
ollama run hf.co/KOHENOOR-AI/kai-lite2:Q4_K_M
```
Unsloth Desktop

How to use KOHENOOR-AI/kai-lite2 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf KOHENOOR-AI/kai-lite2: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": "KOHENOOR-AI/kai-lite2:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use KOHENOOR-AI/kai-lite2 with Docker Model Runner:
```
docker model run hf.co/KOHENOOR-AI/kai-lite2:Q4_K_M
```

Lemonade

How to use KOHENOOR-AI/kai-lite2 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull KOHENOOR-AI/kai-lite2:Q4_K_M

Run and chat with the model

lemonade run user.kai-lite2-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use KOHENOOR-AI/kai-lite2 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf KOHENOOR-AI/kai-lite2: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 KOHENOOR-AI/kai-lite2:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use KOHENOOR-AI/kai-lite2 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf KOHENOOR-AI/kai-lite2: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 "KOHENOOR-AI/kai-lite2: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"

KAIPAL Teacher

A gentle AI teacher that fits in your pocket. KAIPAL Teacher explains any topic as simply as it can be explained, through imagination and learning by doing, on an ordinary phone, offline, with no GPU and no powerful processor.

KAIPAL Teacher (kai-lite2) is the second launch of the KAIPAL series by Kohenoor Technologies, made for Education 3.0 for beginners. About 2 billion parameters in a 1.2 GB file.

Key features

Runs on mobile phones. It needs only about 1.5 GB of RAM and works fine on an average phone from the last three years.
No GPU or powerful CPU needed. It runs on an ordinary processor, offline and private.
Old graphics cards work too. It runs smoothly on an NVIDIA GTX 970 with 4 GB.
Teaches through imagination. Every answer opens with a small scene from everyday life that you step into.
Learning by doing. Every answer ends with "Your Turn! 🌱": one small, safe task to try yourself.
Simple, always. It explains as if to a curious five-year-old, whatever your age: easy words, short sentences, one example, no jargon.
Stays a teacher. Asked for stories, essays, jargon or long technical answers, it keeps teaching, short and simple. It does not play other characters.

How it answers

Every lesson has three parts:

Imagine: a scene from everyday life that you are part of.
Understand: a short, correct explanation in very easy words.
Your Turn! 🌱: one small task to try. It does not give the task's answer.

When you answer the task wrongly, say you are confused, or ask again, it explains with a new scene. Ask about something risky and it says kindly why it is not safe, then gives you a safe task.

Run it with Ollama

ollama run kohenoor/kai-lite2 --think=false

Or from this repository: download kai-lite2-q4_k_m.gguf and Modelfile into one folder and run:

ollama create kai-lite2 -f Modelfile
ollama run kai-lite2 --think=false

Always run it with its Modelfile. The Modelfile holds the teaching instructions and the settings the model was made with. Without them a model this small does not stay disciplined. In a phone app that loads GGUF files, paste the text between SYSTEM """ and """ in the Modelfile as the system prompt. Use temperature 0.3, top-p 0.9, top-k 20, repeat penalty 1.1 and a 4,096-token context. The app needs an inference engine from October 2026 or newer.

Limits of this release

English only.
Text only: it does not read pictures or recordings.
Facts and sums: check what matters. It is a small model and can get a fact or a sum wrong. Teach alongside a grown-up, and check anything you will rely on.
For deeper help, KAI Full is at www.kohenoor.net.

Terms

This release is distributed under the Apache License, Version 2.0. The full text is in LICENSE, and NOTICE states what the files are and that they have been modified. Kohenoor Technologies' own terms for KAI products apply in addition. KAI, KAIPAL and KAIPAL Teacher are names of Kohenoor Technologies.

The checksum of the model file (SHA-256) is in CHECKSUM.txt. The model is also on the Ollama registry at https://ollama.com/kohenoor/kai-lite2.

Downloads last month: -

GGUF

Model size

2B params

Architecture

qwen35

Hardware compatibility

4-bit