Get Started

Public NuGet package family for .NET

UAIX.LmRuntime

Local GGUF and LLaMA runtime packages for .NET, with a managed CPU application path and explicit backend registration, probing, and capability surfaces.

.NET CLI
dotnet add package UAIX.LmRuntime.LocalEndpoint
Get Started
Browse Packages
View NuGet
  • Local model runtime
  • No provider API
  • No model downloader
  • No telemetry

Required For application integration

UAIX.LmRuntime.LocalEndpoint

Verified local file intake, bounded loading, isolated sessions, and deterministic greedy generation.

Execution
Managed CPU application path
Backends
Registry and diagnostics documented separately
Input
Caller-selected local GGUF bytes
Open the package guide →

Package selection

Choose The Package Required For Your Task

Install LocalEndpoint for the managed CPU application path. Reference a lower layer, backend registration, or native-asset identity only when the host owns that exact responsibility.

Application facade

Required For application integration

LocalEndpoint

The high-level local-only GGUF facade: verified files, bounded loading, isolated sessions, UAIX context evidence, and deterministic greedy generation.

Backend selection

Required For explicit backend registration, probing, selection, and fallback evidence

Acceleration

Explicit backend contracts, registration, capability declarations, local probing, deterministic selection, and visible CPU fallback.

Backend registrations

Required For the package-visible managed CPU backend and explicit CPU fallback identity

Backends.CpuManaged

The explicit managed .NET CPU backend registration, capability declaration, device identity, and no-native-asset compatibility lane.

Required For CUDA backend registration and fail-closed CUDA diagnostics

Backends.Cuda

CUDA backend registration, declared compatibility metadata, runtime identifiers, and fail-closed local probe diagnostics.

Required For DirectML backend registration and fail-closed DirectML diagnostics

Backends.DirectML

DirectML backend registration, declared compatibility metadata, runtime identifiers, and fail-closed local probe diagnostics.

Required For Vulkan backend registration and fail-closed Vulkan diagnostics

Backends.Vulkan

Vulkan backend registration, declared compatibility metadata, runtime identifiers, and fail-closed local probe diagnostics.

Required For ROCm backend registration and fail-closed ROCm diagnostics

Backends.Rocm

ROCm backend registration, declared compatibility metadata, runtime identifiers, and fail-closed local probe diagnostics.

Required For Metal backend registration and fail-closed Metal diagnostics

Backends.Metal

Metal backend registration, declared compatibility metadata, runtime identifiers, and fail-closed local probe diagnostics.

Native asset slots

Required For the separate Windows x64 Tesla K80 compute-capability 3.7 native-asset package slot

Backends.Cuda.LegacyK80.win-x64

RID-specific NuGet package slot for Tesla K80 legacy CUDA native assets under runtimes/win-x64/native/.

Model pipeline

Required For GGUF inspection and validation

Gguf

Bounded GGUF parsing, metadata and tensor catalogs, strict validation, sharding, hashing, and mapped tensor access.

Required For token selection and stop handling

Sampling

Greedy and probability sampling, logit transforms, deterministic random state, top-k/top-p filtering, stop matching, and generation control.

Required For managed CPU math

Kernels.Cpu

Reference, portable-vector, AVX2-aware, half-precision, and quantized CPU kernels with explicit dispatch and parity checks.

Required For tokenizer and chat-template work

Tokenization

GGUF tokenizer metadata, tokenizer factories and engines, special-token handling, chat templates, truncation, safety, and parity tools.

Required For LLaMA graph/session internals

Models.Llama

LLaMA-family configuration, tensor binding, mapped weight sources, reference forward execution, sessions, KV cache, generation, persistence, and parity evidence.

Foundations

Required For runtime-neutral contracts

Abstractions

Stable runtime contracts, request and response models, diagnostics, and governance interfaces.

Required For tensor layout and storage metadata

Tensors

Tensor shapes, data types, GGML storage traits, quantized-block metadata, and reference vector math.

Compare package ownership, backend boundaries, and direct dependencies →

Fast path

Run Your First Model

Create a console app, install LocalEndpoint, supply trusted model identity, load the verified GGUF file, create an isolated session, generate locally, and dispose the session before the model.

terminal 
dotnet new console --name LocalModelSample
cd LocalModelSample
dotnet add package UAIX.LmRuntime.LocalEndpoint

The package does not download a model. Use a local GGUF artifact whose source, license, SHA-256, byte count, tokenizer, and architecture you have reviewed.

  1. 01Create the console app
  2. 02Add LocalEndpoint
  3. 03Obtain SHA-256 and byte count
  4. 04Load the verified model
  5. 05Create an isolated session
  6. 06Generate and dispose resources
Open the complete verified-model example → Read the LocalEndpoint package guide →

Layer ownership

Package Dependency Map

The model pipeline and backend-selection branch remain explicit. Every guide and NuGet page lists exact direct dependencies.

Foundations
Model pipeline
Backend registrations
Native asset slots
Application facade
Important boundaryCpuManaged is selectable now. GPU registration packages declare identity and compatibility metadata, then fail closed until a host proves an executable local adapter. A package reference is not execution proof.

Current public scope

Works Today / Not Claimed

Treat package publication, declared compatibility, successful probing, and executed model evidence as separate states.

Works Today

  • Public NuGet package family
  • .NET 9 package targets
  • Local verified GGUF facade on managed CPU
  • Explicit backend registry, probe, and selection contracts
  • Managed CPU backend availability and visible fallback state
  • Generated API reference and source-grounded examples
  • Deterministic fixture evidence

Not Claimed

  • GPU inference from package installation alone
  • Broad model compatibility
  • Benchmark leadership
  • Model downloads or provider API access
  • Production certification
  • Safety certification
Review current project status →Review evidence and benchmark rules →

Find what you need

Start with a task, then go deeper.

Package Family Choose the correct UAIX.LmRuntime NuGet package and understand direct dependency boundaries. Getting Started Install the local facade, verify a GGUF artifact, create a session, and generate within explicit limits. LocalEndpoint Guide Use the application integration facade for verified local managed GGUF inference. Acceleration Guide Register, probe, select, and diagnose local execution backends without silent fallback. Backend Guides Inspect managed CPU and GPU registration package boundaries with direct NuGet links. API Reference Search every documented type and member generated from public package XML documentation. Documentation Task-oriented package, setup, architecture, evidence, security, and terminology guides. Overview What UAIX.LmRuntime is, who it is for, and which responsibilities stay with the host. Project Status Current public package availability, documented boundaries, and evidence posture. Architecture How model-pipeline packages, acceleration contracts, backend registrations, native-asset identities, and the facade fit together. Runtime Capabilities Separate represented APIs, executed evidence, compatibility scope, and public support claims. Evidence & Benchmarks How to interpret validation, parity, real-model evidence, and benchmark claims. Discovery Policy Canonical routes, machine-readable files, metadata parity, support boundaries, and no-action behavior. Package Answers and Source Map Direct package answers, canonical citation routes, structured data, route inventory, sitemap, and search metadata. Security Local file trust, hostile-input boundaries, disclosure, and hosting checks. Glossary Plain-language definitions for runtime, package, evidence, tensor, tokenization, and governance terms. About Why the project exists and the engineering background behind it. Contact Direct contact, corrections, collaboration, and responsible disclosure paths.

Start with the package that owns your task.

Install LocalEndpoint for managed CPU application integration, or choose a lower-level package for backend selection, backend registration, native assets, GGUF, tokenization, sampling, CPU kernels, tensors, contracts, or direct LLaMA execution.

Get Started
Browse package guides

Useful to your audience? Share or link to the specific source.

UAIX.LmRuntime — Local GGUF and LLaMA runtime packages for .NET with managed CPU execution and explicit backend contracts.