lm15 for Java (early port)

lm15 is one request and response model for every major AI model provider: write a request once and send it to OpenAI, Anthropic, Gemini, xAI, Groq, DeepSeek, OpenRouter, a cloud or a local model, by changing the model string. Each language implements it separately and is graded by one shared contract.

Early port, not published. Written against the lm15 contract of 2026-09-11 (cfed007, in CONTRACT_PIN), where it passed every check: 1,380 of 1,380 in 16 directions, plus 165 of its own tests, on Java 21 and 25. It has not been updated since, and the contract has grown (sign-in, judgments, ordered JSON checks, new providers). No provider has been called from this port.

lm15 is released in Python (1.0.1, stable), TypeScript, Rust and Go (release candidates). Guides: lm15.dev. For production, use one of those.

What's here

Synchronous, zero runtime dependencies (the JDK's java.net.http for HTTP and WebSocket), Java 21 or newer: the router, the four provider dialects, streaming, tools, errors, credentials and cloud credential chains, files, batches, caches, media generation, video and realtime sessions.

Try it

mvn -B package        # unit tests + target/lm15.jar

LMRouter router = new LMRouter();   // keys from the environment
Request request = Request.builder("anthropic:claude-haiku-4-5")
    .user("What eats acorns at night?")
    .build();
System.out.println(router.complete(request).text());

Conformance

python3 tools/check_contract.py --contract ../lm15-contract --direction all

The full guide written with this port, with every API and its stated differences from the Python reference: docs/guide.md.

License

See LICENSE.