Olleh provides an Ollama-compatible API to Apple's new Foundation Models, announced at WWDC 2025. It serves as a bridge between Apple's native AI capabilities and the Ollama ecosystem, offering both a command-line interface and an HTTP API for seamless integration with existing tools and workflows.

• macOS 26 beta or later
• Apple Silicon Mac (M1 or later)
• Xcode 26 beta / Swift 6.2+

brew install mattt/tap/olleh

git clone https://github.com/mattt/olleh.git
cd olleh
make
sudo make install # installs to /usr/local/bin/

# Check if Foundation Models are available
olleh check

# Start the Ollama-compatible API server
olleh serve

# Chat interactively with the model
olleh run default

❯ olleh
OVERVIEW: Ollama-compatible CLI for Apple Foundation Models

USAGE: olleh <subcommand>

OPTIONS:
  --version               Show the version.
  -h, --help              Show help information.

SUBCOMMANDS:
  serve                   Start olleh
  run                     Run a model interactively
  list                    List models
  show                    Show model information
  check                   Check availability

  See 'olleh help <subcommand>' for detailed help.

Start the Ollama-compatible HTTP API server.

# Default configuration (port 11941)
olleh serve

# Verbose logging
olleh serve --verbose

# Bind to specific host and port
olleh serve --host 0.0.0.0 --port 11434 # default ollama port 

Start an interactive chat session with the model.

$ olleh run default
>>> Enter a message (/? for help)

Use Ctrl+C or type /bye to exit the chat session.

List all available models. Currently returns only the default Foundation Model.

$ olleh list
NAME                     ID             SIZE     MODIFIED
default                                 N/A      2 weeks ago

Show information about a model.

$ olleh show default
  Model
    architecture        foundation
    parameters          3B
    context length      65536
    embedding length    2048
    quantization        2b-qat

  Capabilities
    completion
    tools

  Parameters
    temperature    0.7

  License
    Apple Terms of Use

Verify that Foundation Models are available on your system.

$ olleh check
Foundation Models available

When running olleh serve, the following Ollama-compatible endpoints are available:

• POST /api/generate - Generate text completions
• POST /api/chat - Chat with the model
• GET /api/tags - List available models
• GET /api/show - Show information about a model

You can use Olleh with the Ollama Swift client library:

import Ollama

// Connect to olleh server (default port: 11941)
let client = Client(host: URL("http://localhost:11941")!)

// Generate text using Apple's Foundation Models
let response = try await client.generate(
    model: "default",
    prompt: "Tell me about Swift programming.",
    options: [
        "temperature": 0.7,
        "max_tokens": 100
    ]
)
print(response.response)

# Generate text with streaming
curl http://localhost:11941/api/generate -d '{
  "model": "default",
  "prompt": "Why is the sky blue?",
  "stream": true
}'

# Chat completion
curl http://localhost:11941/api/chat -d '{
  "model": "default",
  "messages": [
    {"role": "user", "content": "Hello, how are you?"}
  ],
}'

Olleh currently supports the lone default model provided by Apple's Foundation Models framework.

Olleh supports loading custom Foundation Models adapters using the --adapter flag:

# Load and run with a custom adapter
olleh run default --adapter /path/to/my_adapter.fmadapter

Foundation Models adapters let you:

• Specialize the model for specific domains or tasks
• Improve accuracy and consistency for your use case
• Add new skills to the base model

See Apple's Foundation Models Adapter documentation for information on training custom adapters.

Future releases may include:

• Support for specialized models as they become available
• Model configuration and fine-tuning options
• Adapter management commands

This project is available under the MIT license. See the LICENSE file for more info.