Crate chat_applefm 

Apple on-device foundation model provider for chat-rs.

Talks to the ~3B-parameter model that ships with Apple Intelligence (macOS 26+) through Apple’s FoundationModels framework. There is no HTTP and no weights file: the OS owns the model; this crate owns the translation.

How it connects

Rust cannot call Swift directly, so the crate embeds a small Swift package (bridge/) exposing a few plain C functions. Requests and responses cross that boundary as JSON strings — the same mental model as an HTTP provider’s wire format, minus the network. The bridge is compiled automatically by build.rs; users only run cargo build.

On non-macOS targets the crate still compiles (with a stub bridge) and reports the model as unavailable.

Binary crates on macOS need one build.rs line so the Swift concurrency runtime resolves at load time:

// build.rs of your binary crate
println!("cargo:rustc-link-arg=-Wl,-rpath,/usr/lib/swift");

Current scope

Text completion (with optional LoRA fine-tune) and availability probing. Tools, structured output, and streaming are rejected at request time for now and arrive in later slices.

use chat_applefm::AppleFMBuilder;

let probe = chat_applefm::availability();
if !probe.available {
    println!("unavailable: {}", probe.reason.as_deref().unwrap_or("?"));
    return Ok(());
}

let client = AppleFMBuilder::new()
    .with_lora("adapters/transcripts.fmadapter") // optional fine-tune
    .build()?; // validates config (e.g. the adapter path) upfront
// System prompts go through Messages like any other provider; the
// provider maps them onto the session's instructions.

LoRA fine-tunes

AppleFMBuilder::with_lora loads a .fmadapter package — a LoRA trained against the on-device base model with Apple’s adapter training toolkit. Adapters are version-locked to the base model: after a macOS update rolls the model, retrain and reship.

See providers/AGENTS.md for the overall provider architecture.

Structs

AppleFMBuilder

Builder for AppleFMClient.

AppleFMClient

Client for the Apple Intelligence on-device foundation model.

Availability

Result of probing whether the on-device Apple model is usable here.

Enums

Sampling

Decoding strategy for the on-device model — the full set FoundationModels exposes via GenerationOptions.SamplingMode.

Functions

availability

Ask the OS whether the Apple Intelligence on-device model can be used.