The Deterministic Knowledge Engine
Your agent does the work. The engine keeps it honest.
DKE is a deterministic reasoning engine your agent connects to over MCP. It records what your agent learns, reads it back, reasons over it, and states the rules that keep its conclusions current — every answer carrying the sources it rests on. You install it; your agent does the rest.
Connect
Attach the engine to your agent
DKE runs as a service your LLM or agent connects to over MCP. Point an MCP client at the DKE endpoint, hand it your access key, and your assistant can write and run DKE Python, record and recall what it learns, and get answers it can trace — all through the same tool interface it already speaks.
Speaks MCP
DKE is served as a small set of standard MCP tools — compile a module, run a program, list and inspect what's stored. Any MCP-capable client can drive it; there's no bespoke SDK to adopt.
Your key, your store
Each account gets an access key and an isolated store. What your agent teaches DKE stays yours — nothing is shared across accounts.
Answers you can trace
Every result comes back with the facts it rested on — a derivation you can follow, not a guess you have to take on faith.
Tells you what to write
A program that doesn't compile is refused whole — nothing is stored — and answered with the line, the column and the form that belongs there. Your client gets that correction as its own field to apply, not prose to parse, so a first draft is usually one edit from running rather than a rewrite.
Get set up
Your dashboard — stores, keys, and what they cost
An account comes with a dashboard. It is where you create the stores your agent writes to, issue the keys it connects with, and top up the LangCoin that pays for the work.
A key reaches one store, at one level: read, read and write, or read, write and remove. Issue one per agent and you can see when each was last used and which clients have presented it — so a key that has been shared is visible rather than assumed. Change a level or revoke a key at any time.
Learn
The published language
Everything you need to program DKE, published and versioned: a tutorial and a reference for each of the three things you deal with — the language, the wire, and the reasoning modules we publish — and the Language Specification, normative beside them. The language is MIT — read it, build against it, hold us to it.
DKE Python
The language you write. Tutorial from your first line to classes, provenance and modules; reference with every verb, type and construct, worked examples included.
DKE MCP
The wire you send it over. Tutorial for getting a key, the handshake and your first call; reference for the full tool surface, response envelope and error model.
Reasoning modules
Modules we publish, which you compile into your store instead of writing the same declarations again. Tutorial for putting one to work on your own facts; reference for the kinds, derived fields and queries each one brings.
The Language Specification is the normative text: what the language is, in the form you hold us to. The reference above describes the same language and stands on its own authority.