Apfel

What it does

Apfel is a command-line interface to Apple's on-device language model that ships with macOS Tahoe. It provides CLI access to a 3 billion parameter model without requiring separate downloads, API keys, or configuration. The tool installs via a single brew command and runs entirely on the user's machine.

Three interfaces are available from one installation: a Unix-style CLI tool, an OpenAI-compatible server, and an interactive chat mode. The CLI supports piping, JSON output with schema validation, file attachments, and multi-turn conversations. The server runs locally on port 11434 and accepts standard OpenAI SDK clients. The chat mode handles multi-turn conversations with automatic context management and system prompts.

Apfel integrates with the Model Context Protocol, allowing users to attach tool servers that give the model access to calculators, APIs, databases, and custom functionality. Tool discovery and execution happen automatically.

Who it is for

Developers and technical users on Apple Silicon Macs running macOS Tahoe with Apple Intelligence enabled. The tool targets those who want local AI without managing model weights, environment variables, or token costs. It suits users building automation scripts, command-line tools, or applications that need embedded language model capability without external dependencies.

Pricing

Free. The tool is open source under the MIT license.

How it stands out

Apfel eliminates the setup friction that plagues other local AI tools. Most require downloading multi-gigabyte model files and configuring YAML or environment variables. Apfel uses a model already present on supported Macs, installing in seconds with zero configuration steps. There are no token costs, subscriptions, or per-request billing. All computation happens locally with no network calls.

The tool is designed for shell integration rather than as an isolated application. It outputs JSON, respects Unix conventions with proper exit codes, and composes with standard tools like jq and xargs. The OpenAI-compatible server mode allows drop-in replacement of cloud APIs in existing applications.

What a founder should check

First, verify the actual addressable market. Apple Intelligence is limited to recent hardware and specific macOS versions. Check whether potential users will have the required combination and adoption rates over time.

Second, understand Apple's stance on this use case. The company could change the availability of the underlying model, licensing terms, or distribution method. Reliance on built-in OS models creates dependency on a single vendor's roadmap.

Third, examine the moat against cloud alternatives. As inference costs fall and cloud providers optimize for local-first deployment, the primary advantage—zero setup—diminishes. Consider whether lock-in exists and whether users have switching costs beyond convenience.