ai-completion

Tiny shell helper that lets you draft, tweak, and run AI-generated commands straight from Bash or Zsh.

It exposes two public functions:

• ai – talks to the OpenAI Chat Completions API
• aio – talks to apfel (Apple Intelligence) if installed, otherwise falls back to a local Ollama daemon

Quick install

1. Save the script or clone this repo somewhere.
2. In ~/.bashrc or ~/.zshrc add

   source /path/to/ai-completion.sh

3. For ai export OPENAI_API_KEY=<your-key>.

Usage

Describe the task in plain English:

➜ ai organize all photos in this directory into subfolders using the iso date of the image capture time
$ exiftool '-Directory<DateTimeOriginal' -d '%Y-%m-%d' .

The suggestion appears inline on your prompt. Edit if necessary, press ↵, and it executes.

Piping

Anything you pipe into the function is forwarded to the model as extra context, e.g.

➜ git diff --staged | ai write a commit message
$ git commit -am "Refactor payload construction using jq for better JSON handling"

Local

Save a buck using a local model for the easy stuff. Local models can be inconsistent, double check all commands.

With apfel (macOS only, uses Apple Intelligence on-device):

➜ aio list stopped docker containers
$ docker ps -a --filter status=exited

If apfel is not installed, aio falls back to Ollama:

➜ aio list stopped docker containers
Using gemma3:4b on http://localhost:11434...
$ docker ps -a --filter status=exited

Environment variables

Required for ai:

• OPENAI_API_KEY – your OpenAI key

Optional (defaults):

• OPENAI_MODEL (gpt-5-chat-latest)
• OPENAI_TEMP (0.3)
• OLLAMA_MODEL (gemma3:4b)
• OLLAMA_TEMP (0.3)
• OLLAMA_HOST (http://localhost:11434)

Dependencies: curl, jq. Optionally apfel (brew install Arthur-Ficial/tap/apfel) for aio on macOS, or a running Ollama daemon as a fallback.