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, incomplete, not published. Written against the lm15
contract of 2026-09-11 (cfed007, in CONTRACT_PIN). At that commit it
passed the serialization (115), token (43) and credential (43) directions;
the provider dialects, streaming and the other surfaces were not written.
It has not been updated since.
lm15 is released and stable (2.0) in Python, TypeScript, Rust, Go and R. Guides: lm15.dev. For production, use one of those.
.NET 8, zero runtime dependencies: every canonical type and its exact JSON, the error hierarchy, credentials (policies, stores, cloud chains, OAuth, the doctor), SigV4 and RS256 signing, compat presets, the provider registry, transport and SSE. Not yet: the provider dialects, stream assembly and the surfaces that use them. PORTING.md is the map for whoever continues it.
dotnet build -c Release
dotnet testpython3 harness/check.py --shim dotnet --direction serde --no-check-pin # from lm15-contract, with the shim registeredThe full guide written with this port, with every API and its stated differences from the Python reference: docs/guide.md.