aigen

A Swift CLI tool that sends prompts to Apple Intelligence Foundation Models. Reads input from files, inline prompt text, or stdin, optionally with system instructions, and prints the model's response.

Requirements

• macOS 26+ (Tahoe)
• Apple Silicon Mac (M1 or later)
• Apple Intelligence enabled in System Settings
• Swift 6.0+

Installation

Build from source using Swift Package Manager:

git clone https://github.com/kristopherjohnson/aigen.git
cd aigen
swift build -c release
cp .build/release/aigen /usr/local/bin/  # may need 'sudo'

Usage

USAGE: aigen [options] [file ...]

ARGUMENTS:
  file                    Input files to read (use '-' to read stdin; reads stdin if no arguments)

OPTIONS:
  -i, --instruction TEXT  Set system instructions for model (can be repeated)
  -p, --prompt TEXT       Add inline text to prompt (can be repeated)
  -t, --temperature VALUE Set sampling temperature 0.0-1.0 (default: system default)
      --no-stream         Disable streaming; wait for complete response
  -v, --verbose           Show processing details
  -h, --help              Show help information

Examples

Single file:

aigen prompt.txt

Multiple files (concatenated):

aigen system.txt context.txt question.txt

From stdin:

echo "What is 2+2?" | aigen

Pipeline usage:

cat document.md | aigen > summary.txt

Inline prompt text:

aigen -p "What is the capital of France?"

Mix prompt text with files:

aigen -p "Summarize this document:" report.txt

Multiple prompt texts:

aigen -p "You are a helpful assistant." -p "What is 2+2?"

System instructions:

aigen -i "Be concise" -p "What is 2+2?"

Multiple instructions (concatenated):

aigen -i "You are a helpful assistant." -i "Use bullet points." document.txt

Instructions with prompts and files:

aigen -i "Summarize in 5 sentences" -p "Key points:" report.txt

Verbose output:

aigen -v prompt.txt

Non-streaming mode:

aigen --no-stream -p "What is 2+2?"

Custom temperature:

# Lower temperature (0.0-0.3) for more focused, deterministic responses
aigen -t 0.2 -p "What is the capital of France?"

# Higher temperature (0.7-1.0) for more creative, varied responses
aigen -t 0.8 -p "Write a creative story opening"

# Mid-range temperature (0.4-0.6) for balanced responses
aigen -t 0.5 -p "Explain quantum computing"

Explicit stdin with - (Unix convention):

# Append stdin after prompt text
man otool | aigen -p "Summarize this for me:" -

# Mix stdin with files in specific order
aigen -p "Context:" intro.txt - question.txt

# Use stdin at beginning
echo "test input" | aigen - -p "Is this correct?"

How It Works

1. Reads input from inline prompt texts (-p), files, or stdin
2. Optionally sets system instructions (-i) for the model
3. Concatenates prompt texts and file contents with newlines
4. Sends the combined prompt to Apple's on-device Foundation Model with instructions
5. Streams the model's response to stdout in real-time as it's generated (or waits for complete response with --no-stream)

Notes:

• When using both -p and file arguments, all prompt texts are concatenated first, followed by all file contents
• Use - as a file argument to read stdin at a specific position in the input sequence
• If - appears multiple times, stdin is only read once (subsequent - are ignored)
• Multiple instructions are concatenated with newlines and passed to the LanguageModelSession
• By default, output is streamed incrementally so you see the response as it's being generated
• Use --no-stream to wait for the complete response before printing (non-streaming mode)

Error Handling

The tool provides clear error messages for common issues:

• File not found
• File read permission errors
• Apple Intelligence not enabled
• Device not eligible for Apple Intelligence
• Model not ready

Exit codes:

• 0 - Success
• 1 - Validation or runtime error

Development

Run tests:

swift test

Build for debugging:

swift build

Run directly:

swift run aigen prompt.txt

License

MIT License - see LICENSE file for details.