Your firm's own AI workspace
Bee Flow gives your firm its own AI chat: staff ask, write and search in one place that belongs to the firm, on a server you choose and with the model you choose. ChatGPT has the more polished chat; here the firm decides where client data goes.
Plan a call Try the live demoThe chat feels the same to your staff. What changes is where the text goes.
| Your firm's own chat | Public chat in the browser | |
|---|---|---|
| Where conversations are kept | On your own server, or on Bee Flow Cloud in the EU | With the chat provider |
| A chat you do not want kept | Incognito chat keeps no conversation, title or memory. Usage and cost rows, guardrail metadata and the tool log remain | Depends on the provider's settings |
| Which AI model | Nine hosted providers or a model on your own hardware. An administrator sets the model behind each level | The provider's own |
| Personal data | With the guard service installed: detected and blocked before a message leaves (Free), or replaced before the model sees it and put back in the answer (Pro). Detection can miss | Whatever staff paste goes along |
| Your documents | Knowledge bases per team, also from a Nextcloud, OneDrive or Google Drive folder. Answers from a knowledge base cite their source | Uploaded chat by chat |
| Who has access | Your users and groups. Sign-in with Microsoft or Google from Starter | Personal accounts |
For your staff
A chat like the one they already know, in a place that belongs to the firm.
Staff ask a question and get an answer from the firm's own manuals, memos and templates. An answer from a knowledge base cites the document, page or row it used. Knowledge bases are in the Free licence.
An assistant has its own instructions, documents and tools, for example for a first draft of a letter to a client. The product calls them agents. You build them without code, in the Free licence.
An assistant works with Outlook, Gmail, OneDrive, Google Drive or Nextcloud through the account of the person who asks. It sees what that person may see. The built-in integrations are in the Free licence.
For a question you do not want kept. An incognito chat keeps no conversation, title or memory. Usage and cost rows, guardrail metadata and the tool log remain. In the web app.
For the administrator
You set the rules once and they hold for everyone.
For each connected app and each action: reading runs directly, writing can ask first, sending always asks. In the Free licence.
One switch makes every level use a model that is hosted in the EU. The switch covers AI models. Connected apps keep their own settings.
Staff choose Fast, Think, Write or Deep Thinking. The administrator decides which model sits behind each level and can limit a group to certain levels.
Staff can see, edit and delete what an assistant remembers, or switch memory off. The organisation can switch it off for everyone. Government identity numbers, bank and card numbers and passwords are never stored in memory.
Privacy Shield detects 21 kinds of personal data and can block a message (Free). Replacing values before the model sees them needs the Pro licence. Both need the guard service installed, and detection can miss.
A monthly spend limit per person or per member of a group. When it is used up, the next request is refused with a clear message. On a self-hosted server only.
Ask a question about the sample documents. The answer lists the documents it used; click one to read the passage it came from.
Ask the sample documents
Sample data. Nothing you do here is saved.
Honest about limits
ChatGPT has the more polished chat, and a company whose whole product is the chat ships new chat features to its own app first. Bee Flow is about where your client data goes.
A model on your own hardware keeps text inside your network, but needs a capable server. We advise on the size before you buy.
Privacy Shield finds most personal data, not all of it. Clear rules for your team stay important.
There is no iOS app. The web app works in the browser on a phone.
The licence is per user per month: Free for chat and knowledge bases, paid licences from 5 users. Setting it up has a fixed price, agreed before building starts, and management is a monthly amount. Hosting and the use of a hosted AI model come on top; a model on your own hardware has no usage costs.
See the plansForward this page. The facts they will ask for are written down, including what is missing.
Prompts, files and answers: where they go, who can reach them, which model providers are involved and what is logged.
Sign-in, roles, retention and encryption, and what is not covered.
One command with Docker on amd64, or Bee Flow Cloud in the EU.
FAQ
Yes. Bee Flow has an OpenAI adapter and an Azure OpenAI adapter; you bring your own API key. The conversations and the documents stay in your own environment.
Yes. With a self-hosted install and a model on your own hardware, text does not leave your network.
Almost. An incognito chat keeps no conversation, no title and no memory, and it ends when you leave the chat. What remains are usage and cost rows, guardrail metadata and the tool log, and the knowledge search log where that is switched on.
Yes. Chat, assistants and knowledge bases are in the Free licence, which needs no licence key on your own server. Paid licences add things like sign-in with Microsoft or Google, notebooks and team projects. The plans page lists each licence.
ChatGPT has the more polished chat. In Bee Flow you can use the same model families with your own API key, and the firm decides where client data goes.
That is what Privacy Shield is for. With the guard service installed and switched on it checks messages on the way out and blocks personal data, in every licence. From Pro it can replace that data with placeholders that are restored in the answer. Detection is machine learning and can miss.
No. Bee Flow also runs on Bee Flow Cloud in the EU. Which fits, we discuss in a short call.
Read more
The NBA and NOREA AI guide, condition by condition, next to the setting in Bee Flow.
Five questions to ask any vendor of a private ChatGPT for business.
What happens to what you type, and what to agree with your team.
How answers cite their source, and how to set one up.
Where each one fits, side by side.
How your data is protected, without jargon.
Tell us who would use it, which documents it should know and where your data may go. 20 minutes. An honest go or no-go, and what a fixed price would look like.
Plan a call Try the live demo