What is Kisuno?
Kisuno is an AI chat platform for companies that runs on your own server. This page tells you in a few minutes what the system is, what it consists of, what it can do and what it deliberately cannot do. It is written for IT decision makers who need to decide whether to adopt the platform.
Who the platform is for
Section titled “Who the platform is for”There are three starting points. They differ in who runs the server, not in what the platform can do.
| Your situation | Who runs the server | What you do |
|---|---|---|
| You use the platform yourself | your own IT team | set up the server, install the instance, invite your team members |
| You are an SMB customer and have it run for you | your IT service provider | you receive a ready instance and work in the chat and in the administration |
| You are an IT service provider | you, for several customers | one separate instance per customer, on a dedicated or a shared machine |
The third case is the reason behind many design decisions. A customer is always a separate, fully isolated instance with its own database, its own files and its own inference key. More on that on the architecture page.
What the system consists of
Section titled “What the system consists of”The chat interface is LibreChat, an open source platform under the MIT license. Chat, agents, file handling and model connectivity come from there, meaning the application your team members see. The fork deliberately stays close to the original so that updates remain possible.
We added everything a company needs to run it under control: administration of users, roles and permissions, a permission check with logging before every access to a third party system, a usage overview with budgets, a chat pane inside Outlook and an installation package for your own server.
What the platform can do
Section titled “What the platform can do”After installing you get, among other things: chat with streaming and history, a model picker with curated cards, uploading files as context, image understanding on the cards that support it, artifacts such as diagrams and HTML directly in the chat, agents, the scaffolding for skills as standard procedures whose content you create yourself, a long term memory with opt-out, and full text search across earlier conversations.
Three features are deliberately switched off in the delivered state, because they need additional building blocks and would otherwise result in a dead button:
| Feature | Why it is off | What is needed |
|---|---|---|
| file search across uploaded documents | needs additional services | enable it during setup |
| code execution, for example generating an Excel file | needs a code backend | enable it during setup |
| parameter panel for temperature and similar settings | unnecessary for non developers | can be re-enabled per role |
What the platform deliberately does not do
Section titled “What the platform deliberately does not do”- No multi tenancy inside the application. One customer is one instance. This is a deliberate decision, because it puts the separation below the application.
- Users do not create their own connectors. Which third party systems are reachable is decided by the administration alone.
- The bundled example connectors only read. They ship no tools that change anything in the third party system. A tool that does not exist cannot be called by accident.
- The Outlook add-in never sends on its own. A draft lands in the normal reply window, and sending stays with the human.
What open source means here
Section titled “What open source means here”The overall product will be released under the MIT license, the repository follows shortly. In practice that means:
- No per seat gating. The number of users is nowhere limited technically and nowhere capped by the license.
- No feature gating. There is no enterprise edition with extra features. What is in the repository is the full feature set.
- No forced branding. Name, logo and welcome text are configuration.
- Self hosting is free. You pay for operation, support and integration if you want to buy those from someone.
The official LibreChat admin panel is released under the AGPL and is therefore deliberately not used. Instead there is a separate, permissively licensed administration. This keeps the overall product free of copyleft obligations.
Where the data lives
Section titled “Where the data lives”Conversations, messages, users and usage data live in the database of your own instance. Uploaded files live next to it on the same server. There is no back channel to us, telemetry is switched off.
For the model responses themselves the content does leave the server, because the computation happens at the inference provider. The default is EuRouter, where processing takes place in the EU. OpenRouter can be enabled optionally and is not EU resident. Model cards outside the EU are visibly labeled in the interface.
There is a hosted instance in Switzerland. Where your own instance is located is decided by your choice of server.
Scope and boundaries
Section titled “Scope and boundaries”So that you can match the feature set against your requirements, here are the areas that are not part of the delivery or that call for a decision during setup.
| Area | Status |
|---|---|
| web search | The search engine runs inside the instance. You choose the scraper according to your data residency requirements and configure it during setup |
| outbound connections from code execution | are decided during setup. Code execution is switched off out of the box |
| user provisioning via SCIM | not included. Accounts are created on first sign-in via OIDC, revoking access runs through the identity provider |
| automatic billing | receipts as CSV and PDF are included, a payment integration is not |
| OCR, image generation, speech input and output | not included. Scanned PDF files are not read as text |