The method is not a sales argument. It is code you can read.
The four guarantees behind my method are not commercial promises: they are written into an execution engine whose code is public. You can download it, read it, and have your own IT people audit it.
MIT licence · no account needed to read it
The four guarantees, in the code
-
Nothing goes out without a signature
The engine separates what is prepared from what goes out. An automation cannot produce an outbound action until a named human has signed. This is not an instruction someone can forget: it is the order of operations.
-
Your contract outranks the tool
Every component declares what it needs. The engine crosses that declaration with the scope you signed and keeps only the intersection. A component can never grant itself more rights than your contract.
-
Nothing leaves your folder
Directory traversal, absolute paths and symbolic links are refused. The log and the stop button are out of reach of the automations themselves.
-
The log is chained
Deleting or altering a line breaks the chain, and verification reports it immediately. Failures appear alongside successes: a log with no failures is a doctored log.
Three examples, from the same farm
This is the only thing I sell: you see what the agent proposed, what it actually did, and who signed off on it.
What runs, and where
Three components, named. None of them is a service I resell to you: two are open source, the third is free software you install on your own machine.
-
E-NKI-Engine
The code I wrote: scope, signature, chained log. MIT licence, public repository, handed over with the mandate.
-
Hermes — Nous Research
Open-weights models. I build on them without being tied to anyone: the engine depends on no particular provider, and the model can be swapped without rewriting your automations.
-
Obsidian
Your notes and the agent's conclusions are text files on your disk, readable with any editor. Free for professional use. The day you stop everything, you keep the files.
Two modes, and what leaves in each
The difference is not technical, it is legal. Better to settle it before signing.
-
Fully local
Nothing leaves your machineThe model runs on your hardware. No text is transmitted to anyone. It is slower and the model is smaller — but it is the only mode in which "your data does not leave" is literally true. It is the default as soon as client or patient data is involved.
-
Remote model
The text submitted to the model leaves · your files stay with youThe model runs at a provider. Your memory and your documents stay on your disk, but whatever is submitted to the model for processing is transmitted. That requires a processing agreement and appears in your register. I will never describe this mode as "local".
What you own
- Your instance configuration: scope, data sources, settings
- The complete log of everything prepared, executed and approved
- The code of the engine and the components, under a free licence
- No proprietary format, no resale licence, no dependency on me
What it does not do yet
A provider with nothing but strengths is not credible. Here is the real state of things, today.
- The components read files, not yet your business software. Each real connection is built during scoping, one at a time.
- The signature carries a person's name, not yet a personal cryptographic key. The log is verifiable; the signer's identity rests on your organisation.
- This is not self-service software. It is installed and operated under a mandate, with a written scope.
Why make all this public? Because a guarantee you cannot verify is not a guarantee. If you are hesitating between two providers, ask to see the other one's code.
The assessment is free and commits you to nothing.
Forty-five minutes, a written map of what can be automated in your business, and a number. What you do with it is up to you.