LM-Kit One puts private AI tools inside the organisation’s own walls

LM-Kit has launched LM-Kit One, a private AI application server for Windows, Linux and macOS that lets organisations run document intelligence, knowledge search and agents on infrastructure they control. The practical pitch is simple: use current AI tools without automatically sending company documents and prompts to an external service.

Watch Desk analysis

What happened

LM-Kit One combines model execution, document processing, retrieval, agents and operational controls in one deployable product. Its Search → Answers endpoint is designed to return answers grounded in an organisation’s own document collections, with citations. Administrators can control models, access policies and agent permissions, while logs and metrics can remain local, according to LM-Kit’s announcement.

The server supports REST endpoints, OpenAI-, Anthropic- and Ollama-compatible interfaces, plus the Model Context Protocol. It can run on one machine or expand across multiple nodes. Evaluation and development are free with no time limit. Production use is also free for businesses meeting LM-Kit’s stated limits of under US$1m annual revenue, 10 employees or fewer and no more than US$3m in outside capital. Other commercial organisations require a Professional License.

Key findings

• Private deployment
  Models, documents, prompts, results and usage telemetry can stay under the customer’s control when external connections and optional telemetry are disabled.
• Grounded answers
  The Search → Answers endpoint is built to return citation-backed responses from internal document collections.
• Agent controls
  Administrators can set permissions for tools and agents, with local logs and metrics for operations teams.
• Familiar integration paths
  REST, OpenAI-, Anthropic- and Ollama-compatible APIs, and MCP support are intended to reduce integration work.
• A small-business route
  Qualifying smaller companies can use the production product without a licence fee under LM-Kit’s published conditions.

Why it matters

For organisations with sensitive documents, local deployment changes the starting question from “which cloud model should we call?” to “which capabilities can we operate ourselves?” That matters in sectors where data residency, internal access controls or simple institutional nervousness are not minor details.

The more interesting claim is architectural. LM-Kit is packaging models, retrieval, document processing and agent permissions together, rather than asking every team to assemble the stack from several vendors and hope the seams behave. The supplied evidence does not establish independent quality, security, latency or reliability results, so the announcement is a product signal, not a performance verdict.

Our read

LM-Kit One is a sensible answer to a real enterprise headache: useful AI that does not require every document to take a sightseeing trip through someone else’s cloud. It is worth considering for teams that need local control and a shorter route from internal documents to working assistants.

The buying test is less “does it have an agent?” and more “can it produce trustworthy, auditable answers on our documents without creating a new operational cupboard full of mysteries?”

What to watch

• Independent tests of retrieval accuracy, citation quality and agent reliability.
• Security reviews of local deployments, APIs, permissions and tool connections.
• Real-world performance across supported models and multi-node installations.
• How the Professional License treats client projects, redistribution and affiliated companies.

Discussion spark: Would your organisation trust a locally deployed AI server with internal documents, and which evidence would you require before putting it into production?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.