local-code

An offline AI coding assistant for Raspberry Pi 5. Works without internet - perfect for airplanes.

What it does

local-code is a wrapper around aider, which lets you chat with an AI that can:

• Read your code files
• Write and edit files for you
• Suggest shell commands to run
• Auto-commit changes to git

It runs entirely on your Pi using Ollama - no cloud, no API keys, no internet needed.

Quick start

cd ~/your-project
local-code

Then just tell it what you want:

• "add a login page"
• "fix the bug in auth.py"
• "write tests for the utils module"

How it works

┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│  local-code │────▶│    aider    │────▶│   ollama    │
│   (script)  │     │ (AI coding) │     │ (runs LLM)  │
└─────────────┘     └─────────────┘     └─────────────┘

1. Ollama runs the AI model locally on your Pi's CPU
2. Aider talks to Ollama and handles file reading/writing
3. local-code is a simple script that launches aider with the right settings

Commands

local-code                    # Start coding assistant
local-code -m qwen2.5-coder:3b  # Use bigger/smarter model (slower)
local-code --status           # Check if everything is running
local-code --list             # Show available models
local-code --help             # Show all options

Inside aider

Once running, you can:

• /add file.py - Add a file to the conversation
• /drop file.py - Remove a file
• /run npm test - Run a command
• /diff - See pending changes
• /help - All commands
• /exit or Ctrl+C - Quit

Models

Model	Size	Speed	Quality
qwen2.5-coder:1.5b	986 MB	Fast	Good (default)
qwen2.5-coder:3b	1.9 GB	Slower	Better
codegemma:2b	1.6 GB	Medium	Good

Add more models with:

ollama pull <model-name>

Files

~/dev/local-code/
├── lc              # The main script
└── README.md       # This file

~/.local/bin/
└── local-code -> ~/dev/local-code/lc   # Symlink so you can run it anywhere

Troubleshooting

"Ollama not running"

sudo systemctl start ollama

Slow responses

• First response is always slow (model loading)
• Use the 1.5b model (default) for speed
• Responses stream as they generate

Model not found

ollama pull qwen2.5-coder:1.5b

Setup (first time)

1. Install Ollama

curl -fsSL https://ollama.com/install.sh | sh

2. Pull a coding model

ollama pull qwen2.5-coder:1.5b

3. Install uv (Python package manager)

curl -LsSf https://astral.sh/uv/install.sh | sh

4. Install aider

~/.local/bin/uv tool install --python 3.12 aider-chat

5. Create the symlink

mkdir -p ~/.local/bin
ln -sf ~/dev/local-code/lc ~/.local/bin/local-code

6. Add to PATH (if not already)

Add this to your ~/.bashrc or ~/.zshrc:

export PATH="$HOME/.local/bin:$PATH"

Then restart your terminal or run source ~/.bashrc.

7. Test it

local-code --status

Requirements

• Raspberry Pi 5 (8GB recommended) or similar ARM64 Linux
• Ollama
• Aider (via uv)
• ~1-2 GB disk space per model