KE Local Dev Env – Intel Arc / Meteor Lake Coding Agents

This repository contains a small, opinionated setup to run local AI-assisted development workflows on a laptop like:

• Lenovo ThinkBook 16 G7 IML
• Intel® Core™ Ultra 7 155H
• Intel Arc integrated GPU (Meteor Lake-P)
• Intel Meteor Lake NPU (Intel AI Boost)
• Ubuntu Linux

The goal is to:

• Use Ollama with Vulkan (Intel Arc GPU) when possible.
• Expose simple, well-documented shell helpers (ke_*) that coding agents (Claude Code CLI, Gemini CLI, Codex CLI, GitHub Copilot, etc.) can rely on.
• Provide an AGENT.md so agents understand your environment and tools.
• Offer examples for:
  • Local LLM via Ollama
  • Docker + Vulkan (Intel Arc)
  • LM Studio GPU usage
  • CLI coding agents orchestration

---

1. Repository structure

.
├── .ke-dev-env.example     # Shell helpers and environment variables
├── AGENT.md                # Description of the environment for AI agents
├── README.md               # This file
└── examples
    ├── 01-ollama-local-code
    │   └── README.md       # Using ke_local_code + Ollama
    ├── 02-docker-vulkan
    │   ├── Dockerfile      # Minimal Vulkan-capable base image
    │   └── README.md       # How to run containers with GPU
    ├── 03-lmstudio
    │   └── README.md       # LM Studio config for Intel Arc
    └── 04-agents-cli
        └── README.md       # Claude Code, Gemini CLI, Codex + ke_* helpers

---

2. Prerequisites

On the host (Ubuntu):

• Git
• Bash or Zsh
• Docker + Docker Compose
• Ollama (>= 0.13.x)
• Vulkan runtime:
  • mesa-vulkan-drivers
  • vulkan-tools
• Optionally:
  • LM Studio
  • Claude Code CLI
  • Gemini CLI
  • codex-cli / GitHub Copilot

Install Vulkan basics on Ubuntu:

sudo apt update
sudo apt install mesa-vulkan-drivers vulkan-tools
vulkaninfo | grep -i 'deviceName' -m 5

You should see an entry like:

deviceName     : Intel(R) Arc Graphics

---

3. Setup (from git clone)

3.1. Clone and install the local env file

git clone https://github.com/your-user/ke-local-dev-env.git
cd ke-local-dev-env

cp .ke-dev-env.example ~/.ke-dev-env

Now edit ~/.ke-dev-env and adjust at least:

• OLLAMA_MODELS
  The directory where your Ollama models live.
  Example: /media/youruser/DATA/ollama-models

• Any project-specific behaviour inside:

  • ke_run_tests
  • ke_run_app

3.2. Source the env from your shell

For bash:

echo 'source ~/.ke-dev-env' >> ~/.bashrc
source ~/.bashrc

For zsh:

echo 'source ~/.ke-dev-env' >> ~/.zshrc
source ~/.zshrc

---

4. Local quickstart (from a downloaded zip)

If you downloaded a .zip of this repository instead of cloning via Git:

1. Unzip and enter the directory:

   unzip ke-local-dev-env.zip
   cd ke-local-dev-env

2. Copy the example env file:

   cp .ke-dev-env.example ~/.ke-dev-env

3. Edit ~/.ke-dev-env:

   • Set OLLAMA_MODELS to your actual models directory (for example /media/youruser/DATA/ollama-models).
   • Adjust ke_run_tests and ke_run_app to match your usual project commands if needed.
   • Optionally change the default model used by ke_local_code.

4. Hook it into your shell:

   For bash:

   echo 'source ~/.ke-dev-env' >> ~/.bashrc
   source ~/.bashrc

   For zsh:

   echo 'source ~/.ke-dev-env' >> ~/.zshrc
   source ~/.zshrc

5. Open a new terminal and test:

   ke_hw_info        # Show CPU / GPU / NPU summary
   ke_run_tests      # Run tests in a project with pytest
   ke_run_app        # Run your Docker Compose app
   ke_local_code "Explain what this environment does."

If these commands work without errors, your local environment is ready for your coding agents.

---

5. Quick usage

Once the shell has sourced ~/.ke-dev-env, you should have:

ke_hw_info        # Show basic CPU/GPU/NPU info
ke_local_code     # Use a local code-focused LLM via Ollama
ke_run_tests      # Run your test suite (pytest by default)
ke_run_app        # Run your app via docker compose

5.1. Test Ollama with Vulkan

Start the Ollama server:

export OLLAMA_VULKAN=1
ollama serve

Check the logs: you should see a line with a non-zero "total vram" value, indicating the GPU backend is active.

Then, in another terminal:

ke_local_code "Write a small Python function that adds two numbers and explain it."

You can customize the default model used by ke_local_code in ~/.ke-dev-env.

---

6. Examples

Check the examples folder for concrete, documented scenarios:

1. Ollama local code:
   How to use ke_local_code and small code models (e.g. codellama:7b-code) in a local loop.

2. Docker + Vulkan:
   How to run containers that see your Intel Arc GPU through /dev/dri and Vulkan.

3. LM Studio:
   How to configure LM Studio to use Vulkan on Intel Arc, including suggestions for model sizes and quantization.

4. CLI coding agents:
   How to integrate Claude Code, Gemini CLI, and codex-cli with ke_* helpers so they act as “brains in the cloud, hands on your machine”.

---

7. Adapting for other hardware

If you publish this as a template, consider creating branches or variants:

• nvidia-cuda
• amd-rocm
• cpu-only
• intel-npu-openvino

Each variant can keep the same AGENT.md shape but adapt:

• Hardware description
• How to enable GPU/NPU
• Default models and quantization

---

8. License

This project is licensed under the MIT License. See LICENSE for details.

9. Disclaimer

This repository was generated using ChatGPT 5.1 Extended Thinking based on a laptop-class configuration with an Intel® Core™ Ultra 7 155H, Intel Arc integrated GPU (Meteor Lake-P), Intel NPU (AI Boost), and Ubuntu Linux.