LM-Kit One
Deploy a private AI server on your own infrastructure: run models, agents, RAG, search, and document intelligence, compatible with OpenAI, Anthropic, Ollama, and MCP clients.
About LM-Kit One
LM-Kit One is a private AI application server deployed on your own infrastructure, providing model, document, search, and agent capabilities, served externally through OpenAI-, Anthropic-, Ollama-, and MCP-compatible APIs. Existing applications built on OpenAI or other cloud APIs can keep their SDKs and prompts and connect by simply changing the base URL, with no need to rebuild the application stack. The product supports Windows, Linux, and macOS, binds to loopback by default, and supports fully air-gapped deployment. In fully local mode, when external tools and optional telemetry are disabled and model weights are in place, document content, prompts, embeddings, and generated output never leave the host machine; there is also a mode with local processing and external inference, in which source documents remain local while administrators decide which derived results are shared. Pricing: free below the public threshold (annual revenue under 1 million US dollars, no more than 10 employees, and no more than 3 million US dollars in external funding), including commercial use, with no activation and no runtime checks; individuals, educational institutions, nonprofits, and open-source projects are always free. Production use above the threshold requires a Professional license, priced by deployment scope rather than by token or seat.
Best for
Best use cases
Isolated-network or offline environments (such as military facilities and government agencies, where data cannot leave the building—the vendor states LM-Kit runs entirely on local hardware)
Edge deployments like factory production lines, quality inspection stations, and predictive maintenance that need real-time inference and cannot tolerate cloud API network latency
Use cases subject to HIPAA, GDPR, and similar compliance requirements, such as healthcare, where sensitive data must stay on your own infrastructure
Migrating from OpenAI or Ollama: keep your SDK and prompts and only change the base URL
Features
Private AI application server
A server-side deployment on your own infrastructure that delivers model, document, search, and agent capabilities through OpenAI-, Anthropic-, Ollama-, and MCP-compatible APIs.
Compatibility with mainstream client protocols
Point OpenAI-, Anthropic-, Ollama-, and MCP-compatible clients at it without rebuilding your application stack. When migrating from OpenAI, keep your SDK and prompts and simply change the base URL.
Models, agents, RAG, and document intelligence
Once deployed on the server, provides model serving, agents, RAG, search, and document intelligence capabilities.
MCP tool and session isolation
External assistants make session calls through MCP tools; MCP sessions and the files they can access are bound to the authenticated owner, preventing one caller from reading other owners' documents or collections through a shared endpoint.
Policy-controlled outbound access
Networked tools (web search, page fetching, HTTP reads) are governed by an allowlist and outbound patterns, letting operations staff decide what the model or agents can access.
API key authentication
Authentication uses API keys as bearer tokens; servers that allow anonymous local access accept keyless requests from the local machine, while every endpoint in any networked deployment requires a request header.
Chat inference and server-side tools
The chat interface supports streaming events or a single JSON result and can generate multiple candidates; it offers server-side built-in tools (web search, calculator, date/time, unit conversion) and client-delivered tools.
Pros
- Run models, agents, RAG, search, and document intelligence on your own infrastructure, keeping data on-premises
- Compatible with OpenAI, Anthropic, Ollama, and MCP clients—migration only requires changing the base URL
- With local processing plus external tools and optional telemetry disabled, document content, prompts, embeddings, and generated output never leave the host
- MCP sessions and files are isolated by authenticated owner, and outbound networked tools are governed by an allowlist
- Supports Windows, Linux, and macOS, binds to loopback by default, and supports fully offline deployment
- Clear free threshold; individuals, education, nonprofits, and open source are always free
Pricing Plans
Professional
- Required above the thresholds
- Scoped by your order: per deployment for LM-Kit One
- Not metered by tokens or seats
- Carries the assurance a production deployment needs
* Pricing information is for reference only. Please refer to the latest announcement on LM-Kit One's official website.