The same models. A workspace you actually hold

Comparison

Keep the model, change where it runs

Keep the models you use today — Bee Flow ships an OpenAI adapter — and change what surrounds them: conversations encrypted under a key you hold, personal data removed before the provider sees it, and one workspace carrying your automations, your knowledge bases and your meeting notes rather than a chat window on its own. ChatGPT Teams has the better chat experience and gets new model capabilities first; this page is about everything that is not the chat window.

Where ChatGPT Teams is ahead

And it genuinely is

The chat experience itself, and the speed at which new model capabilities land. OpenAI ships to their own product first, and a team that just wants the best chat interface with the newest models will feel that difference. The apps and voice experience are excellent. If your organisation has no constraint on where conversations are processed, that head start is real and this page will not talk you out of it.

What a subscription cannot change

The structural part

In a hosted workspace the model vendor is also the operator: the same company holds the conversations, runs the retrieval and decides the region. That is not a criticism, it is the architecture you are buying. It means the answer to "was this personal data sent to a provider" is a contractual assurance rather than a log entry, and that switching models later means switching products. Bee Flow separates the two — the model is a setting, the workspace is yours.

What changes

Four differences that follow from holding the workspace

Every model, including the one you are using now

OpenAI, Anthropic, Google, Azure OpenAI, Mistral, or any OpenAI-compatible endpoint — including a model running on your own network. Different assistants can use different models, so the expensive one handles the hard work and a local model handles anything touching personal data. Changing your mind later is a dropdown, not a migration.

Conversations encrypted under your own key

Chat and notebook conversations are encrypted with a key derived from the user’s password using Argon2id. Knowledge-base documents and meeting transcripts are stored server-readable, because search and diarisation must read them — we would rather name the boundary than let the strongest claim cover everything.

Personal data removed before the provider sees it

Install the detector and names, addresses, IBANs and national identifiers are tokenised on your own hardware before the prompt leaves, then restored in the answer you read. The model does its job without ever receiving the name.

The chat is one surface, not the whole product

The same workspace runs automations with approval gates, internal apps over your own data model, meeting transcription with speaker identification, and cited answers over your documents — all sharing one permission model and one audit trail. A conversation can trigger a workflow, and both appear in the same trail.

What sits under the chat window

Questions

What people ask before moving off a chat subscription

Can we keep using GPT models?

Yes, and many teams do. Bee Flow ships an OpenAI adapter and an Azure OpenAI adapter; you bring your own API key and the models behave exactly as they do today. What changes is that the conversation history, the retrieval and the audit trail live in your workspace instead of the provider’s.

Is the chat as good?

The chat interface is good and improving, but a company whose entire product is the chat window will keep an edge there — that is an honest read of it. Where we are ahead is everything around the window: your own knowledge base with citations, automations, meeting notes, and the fact that the model is a choice rather than the product.

How do we stop staff pasting customer data into a model?

That is what the detector is for. Once installed it inspects prompts on the way out and, per organisation, will block, redact, ask the user, or tokenise and restore. It runs in a container on your own hardware on CPU, so the inspection itself does not involve a third party.

Do we have to self-host?

No. There is a managed version running in EU data centres if you want the outcome without the operations. Self-hosting is what makes the strongest version of the claim available to you — it is not a requirement for using the product.

What about our existing conversations?

They stay where they are; there is no importer for another vendor’s chat history. In practice teams start fresh, because the conversations worth keeping are usually the documents that came out of them, and those you can upload into a knowledge base on day one.

Point it at the model you already pay for

Bring your API key, run the workspace yourself, and send a prompt with a real customer name in it. Then look at what the provider actually received.