LM-Kit provides a private, self-hosted AI application server and embeddable .NET engine that consolidates the entire AI stack—inference, document processing, retrieval, agents, and fine-tuning—into one installable, versioned product. It enables organizations to run open-weight models entirely within their own infrastructure, eliminating per-token fees and data egress concerns. The platform supports OpenAI, Anthropic, Ollama, and MCP-compatible clients out of the box, with a single installer that replaces a dozen glued-together components.
Funding
Funding not disclosed
Founders
Product
Problem
Organizations face a stark choice between sending sensitive documents and data to hosted AI services—which creates legal, security, and privacy risks for contracts, claims, invoices, and personnel files—or assembling a fragile stack of a dozen open-source components that must be integrated, secured, and maintained separately. This complexity forces teams to either compromise on data governance or dedicate significant engineering resources to build and operate a private AI infrastructure.
Solution
LM-Kit delivers the entire private AI stack as one engine, available as a server you install on your own hardware or as a NuGet package embedded directly into your .NET application. The platform unifies local inference, document processing, extraction, retrieval-augmented generation, agent orchestration, and fine-tuning into a single versioned release, eliminating the need to glue together disparate tools. It speaks multiple API dialects—OpenAI, Anthropic, Ollama, and MCP—so existing clients and command-line tools work unmodified, while a native REST surface exposes advanced capabilities no standard API can express. The server starts fully locked down, with outbound access governed by explicit policy, API keys stored hashed, and all state written to a single configurable directory, making air-gapped operation a supported mode rather than a claim.
Target Audience
Primary customers are engineering teams and enterprises that need to run AI on sensitive documents—contracts, claims, invoices, and personnel files—within their own perimeter, including regulated industries, .NET application developers, and organizations deploying AI to customer machines with offline requirements.
Features
- One installer deploys the complete AI stack—inference, document processing, retrieval, agents, and fine-tuning—as a single versioned release with 150+ updates since 2024, replacing twelve separate components
- Multi-dialect API compatibility (OpenAI, Anthropic, Ollama, MCP) with deep coverage including model management verbs like pull, push, create, and delete, verified end-to-end with stock command-line clients
- Runs on CPU, CUDA, Vulkan, and Metal with no per-token fees, supporting Windows, Linux, and macOS on one or multiple nodes
- Agent Skills framework enables reusable, versioned workflows with hot-reload, guardrails for tool permissions, remote loading from URLs, and portable skill directories that load lazily only when needed
- Enterprise security controls: outbound traffic governed by allowlists and egress modes, API tokens stored hashed, per-model license tracking with named sources, and a single state directory for all server writes
- LM-Kit.NET provides the same engine as an in-process NuGet package, enabling offline AI capabilities directly inside customer applications with zero cloud dependency
- Built-in fine-tuning jobs, multi-agent orchestration with six reasoning strategies, 70+ built-in tools, and full tracing for production observability