An AI-powered oopsie catcher. Bevause typing ish ard and ths hsell is mean.

taipo watches for commands that fail in your shell (command not found) and asks AI to figure out what you meant to type.

If you typo something like:

❯ got status

taipo steps in with:

[taipo] Trying to make sense of: 'got status'... ✔
⚡ Run git status? (y/N):

And if you're in autonomous mode, it runs the fixed command for you:

[taipo] Trying to make sense of: 'got status'... ✔
🚀 Executing git status

• Clone the repo and run the install script:

git clone https://github.com/pg8wood/taipo.git
cd taipo
./install.sh

The installer will prompt you to select your preferred LLM provider, mode, and model settings.

taipo supports multiple LLM backends to fit your workflow. You can run completely locally and privately for free, or connect to cloud services:

Ollama runs high-performance models locally on your own machine. No API fees, no rate limits, and absolute privacy.

• Recommended Models:
  • qwen2.5-coder:14b (Default) - Incredibly fast, highly optimized for CLI and coding.
  • qwen2.5-coder:32b - A great step up for powerful Apple Silicon Macs (like M4 Max with 64GB/128GB RAM).
  • llama3.3 (70B) - Extreme reasoning, perfect if you have plenty of RAM to spare!

Make sure you pull the model first before running:

ollama pull qwen2.5-coder:14b

Use cloud-hosted GPT models. Requires a paid API key.

• Recommended Models: gpt-4o-mini, gpt-4.

You can customize taipo by editing ~/.config/taipo/config.json. Here is a complete example configuration:

{
  "mode": "manual",
  "version": "1.0",
  "provider": "ollama",
  "ollama": {
    "url": "http://localhost:11434/api/chat",
    "model": "qwen2.5-coder:14b"
  },
  "openai": {
    "model": "gpt-4o-mini"
  }
}

For maximum flexibility, you can override any configuration option using environment variables in your shell config (e.g. ~/.zshrc):

• 🟢 manual: You'll be prompted before any suggested command is run.
• 🟣 smart: Asks the AI how confident it is about the typo correction. If 90% confident or greater, the command is run auto-magically. Otherwise, you'll be asked to confirm.
• 🔴 autonomous: Suggestions are executed immediately.

You can change the mode later by editing ~/.config/taipo/config.json or re-running the install script.

Enable debugging by setting an environment variable:

export TAIPO_DEBUG=1

This will print:

• The failed command
• The raw prompt sent to the LLM provider
• The raw response

Helpful for devs or the curious 😎

• Python 3.7+
• zsh
• Local Ollama running OR an OpenAI API Key (set OPENAI_API_KEY in environment)