Assistants, document search and automations, on servers you control

Bee Flow is an AI workspace for teams: assistants, document search with sources and automations in one place. It runs on your own server with Docker, or on Bee Flow Cloud in the EU. Your documents are stored where it runs, and text goes only to the model you choose.

Plan a call Try the live demo

Ask a question, get an answer with its sources

Answers from a knowledge base cite the document, page or row as source chips under the answer; click one to read the passage. [person_1] is a Privacy Shield placeholder: the name was replaced before the text reached the model (placeholders from Pro, with the guard service installed).

Summarise the supplier contracts that renew this quarter

Three supplier contracts renew this quarter. The largest, with [person_1], renews on 1 December and needs notice 60 days ahead. The other two renew automatically unless cancelled.

Sources

Supplier agreement, page 4 · Framework contract, page 2

What is in the workspace

Everything sits behind one navigation, so a person who learns the chat window can find the rest. Each part names the licence it starts in; the plans page has the full list.

Assistants

Assistants without code (from Free)

Build assistants with instructions, tools, knowledge and memory, and share them with the whole organisation or with specific groups. Check an assistant against saved test questions, with a verdict per question (Free). New: an assistant can also start on each new Gmail or Outlook mail, and per assistant you choose what it may do: read only, prepare a draft, or send a reply in the same thread. An incognito chat keeps no conversation, no title and no memory. Usage and cost rows, guardrail metadata and the tool log remain.

Knowledge

Knowledge you can check (from Free)

Add files, text, web pages and tables, or point a knowledge base at a folder in Nextcloud, OneDrive or Google Drive that refreshes when it changes. Bee Flow searches by meaning and by exact words. Answers from a knowledge base cite the document, page or row as source chips; click one to read the passage. Access is set per group.

Card showing a question, the collections an assistant may search and the passages found by meaning and by exact words

Automations

Automations with a brake pedal (from Free)

Start an automation by hand, on a schedule, from a webhook or a form, or on a new mail or another event in a connected app. A dry run reads your real data and simulates every send, create, update and delete, and runs you trust can be replayed after every change (both in Free). Describe an automation in plain language and the assistant builds a draft. It can never switch the automation on: a person does that. Approval steps that pause a run for a person are from Pro.

App Studio

Internal apps from a description (Enterprise)

Describe an internal app, such as a tracker, an intake form or a work queue for a shared mailbox, and the AI builds the tables, screens and logic. You refine it visually, and it runs on the same permissions as the rest of the workspace. App Studio is part of Enterprise.

Meeting notes

Meeting notes (from Pro)

Get a transcript per speaker, a summary, decisions and action items. Upload or record a meeting, or import recordings from Microsoft Teams, Google Meet and Nextcloud Talk.

Models

The model is your choice (from Free)

Nine hosted providers: Anthropic, OpenAI, Google, Google Vertex AI, Azure OpenAI, Mistral, Scaleway, IONOS and EU GPT. Self-hosted models run through Ollama, vLLM, llama.cpp, LM Studio and more. An admin decides which model handles which kind of task.

What an administrator controls

The settings an IT partner looks for first. Where a setting depends on the licence or on self-hosting, the card says so.

What an assistant may do, per action

Per connected app and per action an administrator sets what an assistant may do: reading runs directly, writing can ask first, sending always asks. In every licence.

EU-hosted models only

One switch limits every model level to EU-hosted models. It covers models, not connected apps. A model on your own hardware keeps the text on your network.

Model levels

An administrator decides which model handles which kind of task. On Auto, Bee Flow picks a level per message.

Memory rules

One switch turns memory on or off for the whole organisation, and an administrator can clear all memories. The panel shows counts, never the content.

A monthly AI budget per person

Set how much each person may spend on AI models per month. Self-hosted installs only.

How answers and model choice work

Left: how your documents become answers with sources. Right: how Auto picks a model level for each message. The small classifier step runs when the optional classify service is installed; without it, quick rules and a small AI call decide.

Try the automation builder

Open a step to see its settings, or run a dry run to see what would happen.

Try the automation builder

Sample data. Nothing you do here is saved.

And the smaller parts

Skills

Reusable guidance an assistant pulls in only when the task calls for it. From Starter.

Memory

Assistants remember what matters between conversations. From Free.

Notebooks

Sources on one side, your draft on the other, and a chat about those sources. From Starter.

Projects

Team workspaces with tasks and team chat. From Pro.

Support inbox

A shared inbox for customer support tickets. From Pro, beta.

On a phone

The web app works on a phone. There is no public desktop or Android download and no iOS app.

Connected to the tools you use

More than 50 built-in integrations in every licence, from Outlook and Gmail to AFAS Profit and Frappe CRM. Connecting MCP servers is the MCP server marketplace, on Enterprise and still in beta.

Microsoft 365

Outlook, Calendar, OneDrive and Contacts, including shared mailboxes.

Google Workspace

Gmail, Calendar, Drive, Docs, Sheets and more.

Nextcloud and back office

Files, Talk, Deck and more from Nextcloud, plus AFAS Profit, NMBRS, vPlan and Frappe CRM. AFAS Profit and NMBRS can be read and not changed.

For your IT partner and privacy officer

Forward this page. The facts they will ask for are written down, including what is missing.

Where your data goes

Prompts, files and answers: where they go, who can reach them, which model providers are involved and what is logged.

Technical details per reader

Sign-in, roles, retention, encryption, and what is not covered.

Your own server or Bee Flow Cloud

One command with Docker on amd64, or Bee Flow Cloud in the EU.

Before you decide

Does our data leave our servers?

Text goes to the model you chose, unless that model runs on your own hardware: with a local model through Ollama or vLLM, prompts stay on your network. Calls go to the integrations you connect. By default, a self-hosted install sends nothing to Bee Flow, and the licence check works offline.

Is our content encrypted?

New organisations start with content encryption on managed keys, on every licence. Messages, the extracted text of attachments and transcripts are stored encrypted, except three transcript fields, and an admin can still reset a password. Zero-knowledge encryption (Enterprise), where only the person's own secret unlocks their key, is the stronger option and has exceptions of its own.

Can we start small?

Yes. Free needs no licence key. When you need more, move up a licence on the same install; nothing is reinstalled.

Can we run it ourselves?

Yes. One command installs Bee Flow with Docker on an amd64 server and checks the release before it starts. If you would rather not run servers, use Bee Flow Cloud in the EU.

How much work are updates?

One command pulls the new release, runs the database migrations and checks health. Parts that are still in beta are marked as beta.

See it on your own use case

Tell us about your team and where your data has to live, and we show you which parts fit. 20 minutes. An honest go or no-go, and what a fixed price would look like.

Plan a call See the plans