Serve it
LM-Kit One: the sanctioned translator
One authenticated endpoint replaces the banned tools: intranet forms, helpdesk macros, and pipelines all call the same governed service, and the audit trail shows translation happening inside the walls.
Most companies quietly banned free translators without providing an alternative, so confidential text still gets pasted into them daily. This is the alternative.
Translation demand is constant, the sanctioned tools are absent, and the free ones are one paste away. That gap is a policy on paper and a leak in practice.
One local engine covers the text path and the speech path, so nothing about being multilingual requires being exposed.
02
100+ languages on-device, for snippets, threads, and whole documents.
03
Example pairs teach the engine your glossary, so vocabulary stays yours.
The scenario decides the form: a translation endpoint for the whole company, or translation living inside your own software.
Serve it
One authenticated endpoint replaces the banned tools: intranet forms, helpdesk macros, and pipelines all call the same governed service, and the audit trail shows translation happening inside the walls.
Embed it
Ship in-product translation in your .NET application: document tools, support consoles, field software that works offline, with no per-character cost attached to your feature.
Better than the banned option, not just safer: that is what actually changes behavior.
Private
The text never leaves your infrastructure; verify it with your own firewall rules rather than a vendor's privacy page.
Consistent
Example-guided translation applies your terminology on every pass, so product names and legal phrases stop mutating between documents.
Complete
Documents, threads, and live meetings share the same local pipeline; being multilingual stops being a data-handling event.
One REST call for the company endpoint; one class with a glossary for the embedded form.
The company translation endpoint, callable from any tool or script.
curl http://your-server:5189/lmkit/v1/translate \
-H "Authorization: Bearer your-server-token" \
-H "Content-Type: application/json" \
-d '{"input":"Le contrat prend effet à la signature.",
"language":"English"}'
The embedded engine, with example pairs teaching your terminology.
using LMKit.Model;
using LMKit.Translation;
var model = LM.LoadFromModelID("qwen3.5:4b");
var translator = new TextTranslation(model);
string english = translator.Translate(supportThread, Language.English);
string german = translator.Translate(contractClause, Language.German);
The pipeline composes documented capabilities you can inspect on their own.
Translate
100+ languages with example-guided terminology, entirely on-device.
How translation worksDetect
Text and documents identified across 100+ languages before routing.
How detection worksSpeak
Live translation of meetings and calls while they are transcribed.
How speech translation worksDocuments
PDFs and Office files become translatable text and return to shape.
How conversion worksPolish
Tone and style rewriting keeps translated output in your voice.
How enhancement worksRelated
The speech-side use case this one completes for multilingual teams.
The transcription use caseA ban removes the tool but not the need, so the pasting moves somewhere less visible. A sanctioned endpoint that is faster, respects your terminology, and costs nothing per use replaces the behavior instead of hiding it.
The translation engine covers 100+ languages, with automatic language detection for routing and speech translation for live audio. Quality per language pair depends on the model you choose; evaluate on your real content.
Yes. Example-guided translation takes pairs from your glossary and applies them across every pass, so domain vocabulary stops mutating between documents and translators.
There is no per-character meter. Whether it is one clause or the entire support archive, the cost is your hardware's time, which is what makes company-wide adoption viable.
Private AI, multilingual