An Open WebUI alternative for teams that need more around the chat

Bee Flow is an AI workspace for organisations that you run on your own server, with local or hosted models. Open WebUI is a self-hosted AI platform that does that too. Next to the chat and the knowledge bases, Bee Flow adds automations, a personal-data filter and a Nextcloud connector in one workspace. This page sets them side by side.

Plan a call Try the live demo

For work with client data in an accounting or tax firm, see the page for accounting and tax firms.

Accounting and tax firms

The comparison in one table

Points a buyer can check, with the Open WebUI side taken from its own licence, documentation and repository.

Open WebUI facts from its LICENSE file and from docs.openwebui.com (the license, enterprise, features and quick-start pages). Licence and features checked in November 2026; architecture and Kubernetes last checked on 16 November 2026. Check them there before you decide.

Bee FlowOpen WebUI
LicenceSustainable Use License v1.0, a fair-code licence: the source is public, but it is not open source. Hosting it as a service for others needs an agreement with us. The Nextcloud connector is AGPL-3.0.Open WebUI License from v0.6.6: the branding must stay intact unless you have at most 50 users in 30 days, written permission or an enterprise licence. Its docs say it is not OSI-approved.
Free tierThe Free licence has no time limit and needs no key. It includes chat, knowledge bases, all built-in integrations, the Nextcloud connector, personal automations and, with the guard service installed, blocking of personal data.Free to use with the original branding intact. An enterprise licence adds white-labelling, support and enterprise-only offerings such as Terminals and Tether.
Users and groupsOrganisations, users and groups in the Free licence, with password or Nextcloud sign-in and two-factor authentication. Google and Microsoft sign-in come with Starter. No SAML, SCIM or LDAP.Roles, groups and per-resource permissions. Sign-in through OIDC or LDAP, and provisioning with SCIM 2.0.
Knowledge bases and citationsFiles, web pages, tables, meeting notes and Nextcloud, OneDrive or Google Drive folders. Vector and keyword search, reranked on your own server unless you configure a provider's reranker. Answers from a knowledge base cite the document, page or row.Uploaded files and web pages. Hybrid BM25 and vector search with reranking when you switch it on. Answers cite their sources.
AutomationsMulti-step flows started by a schedule, a signed webhook, an app event or a button, with AI steps, conditions and code, and dry runs that send nothing, in Free. Sharing and approval steps come with Pro.Scheduled prompts: each run starts a new chat with the model's tools, on a recurring schedule.
Personal-data filterPrivacy Shield, with the optional guard service, finds 21 kinds of personal data on your own server and blocks them, in Free. Replacing them before the model and putting them back in the answer comes with Pro. Voice transcripts and builder chats are not scanned yet.We found no built-in filter for personal data in its documentation.
IntegrationsMore than 50 built in, all in the Free licence: Google Workspace, Microsoft 365, Nextcloud, GitHub, YouTrack, web search and more. Further MCP servers are connected through the MCP marketplace, which is Enterprise and still in beta.MCP servers over Streamable HTTP, OpenAPI tool servers, Python tools and functions, web search and image generation.
NextcloudA connector app in the Nextcloud App Store, Nextcloud sign-in, and Nextcloud apps such as Files, Mail, Calendar, Deck and Talk as tools. All in the Free licence.Not covered in its documentation.
Model providers and local runtimesNine hosted providers: Anthropic, OpenAI, Azure OpenAI, Google, Google Vertex AI, Mistral, Scaleway, IONOS and EU GPT, plus Ollama, vLLM, llama.cpp, LM Studio, SGLang or any OpenAI-compatible endpoint.Ollama, OpenAI, Anthropic and any OpenAI-compatible provider. Built to run entirely offline.
Processor architectureImages for amd64 (x86_64) only.x86_64 and ARM64, including Raspberry Pi, according to its quick-start guide.
KubernetesDocker Compose is the supported route. For Kubernetes there is a pattern guide, without a Helm chart.A Helm chart and a Kubernetes deployment guide.
Compliance toolingThe Compliance Center (Enterprise) runs readiness checks for the GDPR, the AI Act, ISO 27001 and DORA, and keeps registers such as DPIAs and incidents. Readiness checks, not a certificate.We found no compliance registers in its documentation.
Who it is built forOrganisations that want assistants, their documents, automations and privacy controls in one workspace they run. Several parts are still in beta.Its documentation says it is built for teams, from a solo install to an organisation with thousands of seats.

Where the difference shows in daily use

Personal data is checked before the model

With the guard service installed, Bee Flow scans chat messages and attachments on your own server and blocks personal data before it reaches a hosted model. If the guard is down, chat messages are blocked rather than sent unscanned. Detection uses a model, so a bare number with no context around it can slip through.

Automations with more than a clock

A flow can start on a Nextcloud or Microsoft 365 event or on a webhook, run an AI step and branch on the result. Every run is logged step by step. From Pro, a step can wait for a colleague to approve it, and meeting notes join the same workspace, with a transcript, decisions and action items.

Compliance work in the product

The Compliance Center, in Enterprise, runs readiness checks and keeps the registers an auditor asks for, with a PDF report, so they do not live in a separate spreadsheet next to your AI tooling. It shows readiness. It is not a certificate.

Nextcloud as the home base

Install the Bee Flow app from the Nextcloud App Store, sign in with your Nextcloud account and let assistants work with files, mail, calendars and Deck boards. A Nextcloud folder can feed a knowledge base directly.

Live demo

A knowledge base, live

The sources a knowledge base keeps up to date: uploaded files and a documentation site read on a weekly schedule, with what came out of each file and what was skipped.

This is the real interface running on sample data. Nothing you do here is saved, and nothing leaves your browser.

Which one to choose

Choose Open WebUI when

You mainly want a good chat on local models, and you want it lighter to install. Open WebUI will serve you well: it installs with pip or Docker, runs offline, runs on ARM as well and you can extend it in Python. Bee Flow is a larger install, with a server, PostgreSQL with pgvector and object storage. It needs 4 GB of memory for the core install and about 11 GB with the personal-data filter and the topic classifier. For a chat on your own machine that is more than you need.

Choose Bee Flow when

Your organisation also needs mail and files under each user's own rights, a personal-data filter, approvals and automations in the same install. Bee Flow puts assistants, knowledge bases, automations and meeting notes in one workspace, with a filter on chat messages before they reach a model. It fits especially well when your organisation already works in Nextcloud.

Questions from teams that run Open WebUI

Is Bee Flow open source?

No. Bee Flow is source-available: the source is public, and you may read it, change it and run it for your own organisation under the Sustainable Use License. That is a fair-code licence and it does not meet the open source definition. Open WebUI's documentation says the same of its own licence from v0.6.6. Of Bee Flow, only the Nextcloud connector is open source, under AGPL-3.0-or-later.

Can we keep our local models?

Yes. Bee Flow works with Ollama, vLLM, llama.cpp, LM Studio, SGLang and other OpenAI-compatible runtimes, so the endpoint you already run can serve Bee Flow as well. You pick the model per assistant and can mix local and hosted models. The usage report counts local models at zero cost.

Can we bring over our Open WebUI chats and knowledge?

There is no importer for Open WebUI chats. Documents move easily: upload the files into a Bee Flow knowledge base, or point one at a Nextcloud, OneDrive or Google Drive folder.

What does the personal-data filter need?

The guard service, an optional container that runs on your own CPU next to the server, without a GPU. Together with the topic classifier it takes about 7 GB of extra memory. Detection combines patterns with checksums and a GLiNER model, and all of it stays on your own hardware. Without the guard service there is no detection of names and addresses.

Do we have to self-host?

No. Bee Flow also runs on Bee Flow Cloud in the EU. The deployment page compares the two options.

Self-hosting an AI workspace with a local model on Ollama

Read more

From a fresh Docker host to a first chat answered by a model on your own hardware: what the install script starts and what runs on CPU.

Read the post

Hybrid search with citations

Read more

How a knowledge base finds the passage: vector and full-text search side by side, reranking on your own server and citations down to the page.

Read the post

Point Bee Flow at the models you already run

Install the free version on your own server and connect your Ollama or vLLM endpoint, or talk your setup through with us first.

Plan a call Install free