Stays with you
The knowledge base in full, the history of every dialogue, customer contacts, the bot's settings and the widget's appearance — all of it in a volume on your own server.
YOUR KNOWLEDGE FINDS ITS VOICE.
Your answers. In every conversation.
Self-hosted AI assistant for your websiteExplore the possibilities Arevian product · in development
01 / YOUR KNOWLEDGE FINDS ITS VOICE.
A good answer starts with your knowledge.
A support chat grounded in your knowledge base, with a manager ready to take over. Runs on your server.
One script tag on the site. Your knowledge base and conversations stay on your server; an external model receives the context needed to answer if you choose one.02 / From zero to a working assistant
Interactive example · not live dataGrounded when it can answer. A person when it cannot.
Choose a question and see what Mimir relies on. This example has delivery rules, but no information about crypto payments.
Your source
Delivery → Gyumri
Saturdays. Orders before 2pm arrive the same day.
Do you deliver to Gyumri on Saturday?
m.
MIMIR
Yes. Order before 2pm for same-day delivery.
Answer found in your source
03 / Architecture and capabilities
AREVIAN / MIMIRWhat the manager gets, not just the visitor
A chatbot is only half the product. The other half is the desk your team works at every morning.
Deploy on your server
Prepare Docker, copy the configuration, set the administrator credentials and start the container. The database, search model and admin panel run there; the answering model is configured separately.
Answers grounded in your material
Semantic search finds relevant topics and the model uses them to answer. If it finds no relevant material, Mimir says it cannot answer reliably and offers a human handoff. As with any AI answer, check important facts before launch.
A co-pilot for the person on shift
Open a conversation and the panel already shows a summary of what the customer wants, the knowledge-base topics that apply, and three ready replies to send or edit.
Take over in one click
The manager takes over in the same chat window and the bot stops replying. The conversation and its context remain in the panel.
Leads instead of lost conversations
Contacts left in the chat become a list with a status, not a stream of messages. The dashboard opens on what is actually waiting today: open leads, escalations, dialogues since morning.
Your brand, down to the CSS
Accent, background, text, position, size, avatar — with a live preview. If that is not enough, there is a field for your own CSS, and the widget carries no framework to fight with.
Keys that nobody can read
The model provider's API key is stored encrypted and is decrypted only in memory at call time. It is masked in the logs, and it never appears in the interface after it is saved.
04 / Under your control
Where your data actually is
This is the question that decides whether a chatbot is allowed near real customers. The answer here is short and checkable.
Leaves only to answer
With a cloud model, the provider receives the current question, relevant excerpts and the conversation context needed for the answer. The full knowledge base and stored conversation history remain on your server. A local model keeps generation in your own environment.
Search runs at home
The semantic search model runs locally on the CPU. Your knowledge base is never uploaded anywhere to be indexed — there is no vendor holding an embedding of your business.
You are the operator
You run Mimir on infrastructure you control and decide who can access it. If you choose an external model or ask us to operate the installation, review those data flows and access terms separately.
05 / How to start
From a working product to a repeatable launch
The repository already contains the widget, admin panel, knowledge base and manager handoff. The next step is to make installation and support dependable for each new customer.
Working now
Self-hosted service, embeddable widget, manual Q&A knowledge base, conversations and manager takeover.
Next
Pilot deployments with real knowledge bases, answer review and clear operating responsibilities.
Later
Easier content import and integrations where customer demand justifies them.
Demo
See answers to questions your visitors actually ask.
- Your five frequent questions
- Answers grounded in your material
- Discuss gaps before launch
Deployment
Install the service, connect the model and embed the widget on your site.
- Docker deployment
- Knowledge base and widget setup
- Manager handoff workflow
Care
Keep the knowledge base, answers and operation under review as your business changes.
- Knowledge base updates
- Review unanswered questions
- Support terms agreed separately
A few more answers.
Can it work without sending anything to the cloud?
Yes. Point it at a local model running on your own hardware and the loop closes entirely inside your network: search is local by default, generation becomes local too.
What does the bot do when it does not know?
When no relevant material is found, it explains that it cannot answer reliably and offers a manager. The conversation appears in the panel. Responses generated from found material should still be reviewed before launch.
How long does it take to launch?
The Docker service can be started after server and credential setup. The larger task is collecting accurate Q&A topics, choosing a model and checking real conversations. We will estimate that work from your material.
Will it fit our design?
Colours, position, size and avatar are settings with a live preview, and there is a field for your own CSS underneath. The widget has no framework of its own, so nothing fights your styles.
Which model should we choose?
Mimir supports GigaChat and OpenAI-compatible endpoints, including local runtimes such as Ollama. Choose based on data policy, answer quality and cost; we will test the selected connection with your knowledge base.
Next step
Let it answer on your site first.
Send us the five questions your support answers most often. We will show you Mimir answering them from your own material before you decide anything.
What happens next
- 01Ari asks three questions
What you're building, what already exists and how to reach you.
- 02The team replies
Within one working day — by email or in Telegram.
- 03Scope and price in writing
Agreed before any work starts.