Outcome 01
Extract
Upload an invoice, a form or a scan, describe the fields you want, and confirm you get JSON matching that schema, with a confidence figure per field and null where the document genuinely does not say.
Quickstart · LM-Kit One
Prove three outcomes in your first hour.
One install, then three things worth checking before you decide anything: a document becomes structured data, a folder becomes searchable knowledge with citations, and a client you already own points at your own server.
What a first evaluation should establish.
An evaluation that only proves the server starts has not told you much. These three cover the layers that matter: data out, knowledge in, and integration.
Outcome 02
Ground
Build a collection from a folder of internal documents, ask a question whose answer lives in one of them, and check that the response carries the document, the page and the passage it came from.
Outcome 03
Connect
Point an existing OpenAI, Ollama or Anthropic client at the server by changing its base URL, or connect an assistant over MCP, and confirm your own tooling works unmodified.
Six steps, in order.
Nothing here needs a database, a container runtime or a second service. The exact commands and settings for your platform come with the Preview package.
- Install the server. It runs as a Windows service or for a desktop session, and on Linux and macOS. A fresh install listens on the local machine only, so nothing is exposed until you decide to expose it.
- Load a model. Pull one from the catalog or import your own. The console shows whether it fits the hardware before you commit to the download.
- Extract from a document. Send a PDF, scan or Office file with the fields you want, and read back schema-matching JSON with per-field confidence.
- Build a collection. Point the server at a folder, let it index, then query by wording, by meaning or both, and check the citations against the originals.
- Connect a client. Change the base URL on an OpenAI, Ollama or Anthropic client, or enable the MCP endpoint and connect an assistant to a curated tool set.
- Set the boundary. Create an access token, decide whether the server stays on loopback or goes on the network, and choose which tools are reachable and what they may return.
When Business Preview is a fit.
The product is built and running, and the boundary is worth stating before you plan around it rather than after.
Good fit
Evaluate, prototype, pilot
Technical evaluations, internal prototypes, controlled pilot deployments, product integration work, and security or architecture review. Preview participants work directly with the engineering team and shape the release.
Wait for GA
When stability is contractual
Support and response-time commitments apply during Preview. What is not yet committed is interface stability across versions and long-term-support builds. If your deployment is immediately mission-critical, or you need that stability in writing before you start, talk to us about timing.
LM-Kit One