Sepp Guard
The assistant works in a sealed room. Only what has been cleared beforehand is allowed.
- It sees the current project and nothing else on the machine.
- No internet and no access to sensitive data without permission.
- The boundary is drawn by the operating system, not by the program.
- Every access is recorded and stays readable afterwards.
sepp mini
one binary
19 MB
Audit trail
Every step is recorded: who did what and when, and what was refused. Not as a memory log, but as a verifiable record.
- What was allowed is on the record too, not only what was denied.
- Delegated subtasks appear indented beneath their parent task.
- The record can be exported and evaluated by machine.
- That makes every run traceable and usable in an audit.
Three ways to run it
One program, three ways in. All that changes is how you address it and where the answer goes.
The gain: control. The course of work stays visible on screen, delicate steps are confirmed beforehand, and you can stop at any time. Work you started can be picked up again the next day.
$ sepp # start the interface $ sepp -c # resume the last session $ sepp -r [id] # resume a particular session
The gain: automation. One question, one answer, and the result goes straight to the next program. That way recurring work runs unattended, overnight as well as in the build.
$ sepp -p "Summarise README.md" $ sepp --provider openai -m gpt-4o-mini -p "…" $ sepp -m kimi-k3 -p "…" # the provider follows from the model
The gain: integration. Requests in, answers out, line by line. The assistant sits inside your own application, without anyone having to open a terminal.
$ echo '{"type":"prompt","text":"hello"}' | sepp --rpc $ sepp --provider local -m llama3 -p "…" # your own machine
Extensions
A plugin is ordinary code, written with the sepp SDK in a language of your choice. It becomes a single building block that sepp runs the same way on every machine: sealed off, with nothing but the rights cleared beforehand.
Any language, SDK included
Plugins as building blocks, platform independent
sepp
runs them, only with cleared rights
The same module everywhere
36 extension examples
Anything that recurs in your organisation and is done by hand today can be taken over by a plugin. Each one gets only the rights it needs for its single task: invoice checking sees the receipts folder and nothing else.
Connecting to your systems
Calculating and checking
Documents and knowledge
Security and evidence
Operations and routine
Which of these fits you? Arrange a call →
Packages: everything in one file, signed
A package bundles skills, templates, rules, plugins and settings into a single file. It is signed with the publisher's key, or with your own, and set up with a single command.
Everything that belongs to one task
One file, signed
sepp pkg install …
One command, ready to use
The fingerprint decides
The manifest inside the package is signed with the publisher's key and names the SHA-256 of every file. An altered package fails here; not a byte reaches the disk before this check.
Confirmed once per publisher. After that every further package must be signed with exactly this key.
An unsigned package is refused: no unpacking, no copy, no rights. The same goes for files altered after the fact.
Packages can come from a directory of your own whose key is registered once. After that the name is enough, and the registered key must still match the package.
Your own marketplace, on your own network
The source becomes a directory for the whole organisation: a signed catalogue of the packages cleared at your site. It sits on your own web space, inside the company network, and needs neither an account nor an outside provider.
Departments and IT put packages together and sign them with their own key.
A signed catalogue on your web space. It names packages; it grants nothing.
Search by name and set up. The registered key must match the package.
sepp pkg search rechnung · sepp pkg install rechnungspruefung
Catalogue and packages stay with you. No outside operator, no dependency.
A folder behind your own web server is enough. The network need not leave the building.
Only what is in the catalogue arrives by name. The operator key is registered for good.
Trust in a publisher can be withdrawn at any time and packages removed.
Which language model? Your decision
sepp does not think for itself, it asks a model. Which one is your decision: a large model from outside, a server on your own network, or a model on the workstation. You switch with one setting at startup; everything else stays the same.
The large outside providers, when performance is what counts. The access key comes from the environment, never appears in any output and is not passed on to subprocesses.
A model you run yourself on your own hardware inside the company network, for instance via LM Studio, Ollama or vLLM. The content never leaves the building, and usage is not billed per request.
On Apple machines the model runs through MLX directly on the Silicon processor, noticeably faster than by the usual routes. sepp finds the running service on its own, without a key and without configuration.
Confidential work stays on your own server, demanding work may go outside, everyday work runs at the desk. Switching is one setting at startup, not a rebuild.
If prices, limits or availability change, you take a different model and carry on. Tools, clearances and records stay as they are.
Questions before the conversation
What prospective customers ask first. Everything else is settled fastest in a call.
What does sepp mini cost?
The price depends on scope and number of workplaces and is agreed in conversation. There is no account, no subscription with an outside service and no charge that scales with your usage.
How does the rollout work?
As a pilot on one of your processes: we set sepp mini up together, your staff work with it, then you decide. Copy one file, make it executable, start it — technically nothing more is needed.
Does it need a cloud provider?
No. You decide which language model answers: a model on your own machine or server, a provider of your choice, or both depending on the task. Your data goes only where you say it goes.
Does it run without internet?
Yes. With a local model, sepp mini works entirely without a network connection — on a notebook, a server on site or from a USB stick, without installation and without admin rights.
Where does the data live?
With you. There is no KIONOVA service that sees your documents. What the assistant reads or writes stays in your environment, and every access is in the audit trail.
What may the assistant do, and what not?
Only what has been cleared beforehand: it sees the current project and nothing else on the machine; no internet and no access to sensitive data without permission. The operating system draws the line, not the program, and every refusal is recorded.
BEHIND THE PROJECT
One binary. One conviction. Sovereignty.
Amir Herco
Visionary & developer
KIONOVA®
„sepp mini is the smallest part of KIONOVA® and at the same time the most honest: an assistant that may do only what you expressly give it. One file, no account, no outside service that fails tomorrow. Anyone who wants to keep their data should not have to choose between progress and control. Developed and built in Germany. And the mini in the name is a statement: I am working towards a general artificial intelligence. When it stands, it will carry the same name, just without the mini.“
Amir Herco, founder and developer · KIONOVA®
Let us talk about it
Whether sepp mini fits your organisation is settled fastest in conversation.
- You write us a few lines about what it is about.
- We get back to you within one working day.
- A 30-minute call about your use case — then you decide.
KIONOVA® is a registered trademark with the German Patent and Trade Mark Office (DPMA), registration number DE 3020251225814.