Knowledge Solution Kit
Answers from your own documents, with sources, for the people allowed to see them.
In development
An engine that finds the passages in your own documents that answer a question, shows exactly where each one came from, and never shows anyone something they aren’t permitted to read. It runs on infrastructure you control.
The questions it answers
Questions whose answers already sit somewhere in your policies, procedures and documentation. It is meant for staff questions like these, not for real-time analytics or for making legal decisions on its own.
Answers that come with their source
Every passage it returns names its source: the document, the version of that document, and the place in it. Anyone can open the original and check before acting on it.
Only what the person asking may see
The engine checks who is asking before it searches, then checks again against your current permissions before anything is returned. A passage from a document someone can no longer open is dropped before it reaches them.
It says so when the answer isn’t there
When your documents don’t contain enough to answer, it reports that plainly instead of producing a confident guess.
Useful without an AI model
Finding and citing passages works with no AI model connected at all. A written answer is an optional step on top, using a model you choose, and every citation in it must point to a passage that was actually found.
No screen of its own
The kit is an engine, not another app to log into. Your website, staff tools or internal systems send it questions, and your developers connect them with the packages that come with it.
Why permissions and sources matter
A small organization’s files mix what everyone may read with what only a few should: personnel matters, board discussions, client records. A search tool that ignores those lines turns a shared drive into a leak, and an answer without a source is an answer nobody can check.
The kit treats both as rules that can’t be switched off. Permissions narrow the search before it starts and are verified again before anything is returned. An AI model never decides who may see what, and text inside a document is treated as content, never as instructions. These protections are part of every license, and they stay in force even if a license lapses.
Every question can be traced afterwards to the sources it drew on. Anything that was withheld is recorded only by reference, never by its text.
Running it on your own infrastructure
Your documents, and everything built from them, stay inside your organization’s boundary.
What you run
The kit arrives as a bundle: the engine in a container, a PostgreSQL database, one configuration file and one license file. It runs anywhere containers run, from a cloud account to a single virtual machine, and it can use a database you already manage instead of the bundled one.
What stays yours
Your content, the indexes built from it and the record of every question stay in your deployment, and you own all of them. The license is checked offline, so the engine never needs to contact us to keep working.
What it asks of you
Someone has to operate it: keep it running, take backups and apply upgrades. The kit comes with health checks, backup and restore guidance and an upgrade procedure. An upgrade replaces the engine, and going back to the previous version is supported.
We don’t offer to host it for you today.
What your developers get
The engine is not delivered as source code, so an upgrade never means merging changes. Your developers connect through its documented web interface, with the JavaScript packages that come with it or any other HTTP client. Anything specific to your organization lives in configuration and in your own code outside the engine.
Where it stands
The Knowledge Solution Kit is in development. Work on its first release began in September 2026, and that release has to find and cite the right passages for a set of test questions, with no AI model connected, before anything is built on top.
The first production release will be shaped by the first organization to use it, and tested against questions its own staff write and approve.
Two limits are worth knowing now. Search in Chinese will be weaker than search in English until a later release adds language-aware search, and scanned documents and tables also come in a later release.
Documents your staff can’t search?
Tell us where your answers live today and who should be able to see them. If the kit fits, your organization could shape its first production release.