Comparison
Google Workspace is our deepest integration
Connect the same Google Workspace — Gmail, Calendar, Drive, Docs, Sheets, Contacts, Meet recordings; it is the deepest integration we ship — and keep the model a setting rather than a property of the suite, Gemini included, through Google AI Studio or Vertex. Inside Docs and Gmail, Google wins on convenience and will keep winning. Beside them you get one workspace that spans more than one vendor’s apps, and a privacy layer sitting between the two.
See the integrations All comparisonsWhere the two differ
Being right there in the document. Nothing beats a sidebar in the file you already have open, and for drafting in Docs or summarising a thread in Gmail the integration is as tight as it gets because Google builds both halves. If your organisation is entirely on Google and the data question is settled, that convenience is real and difficult to match from outside.
Run Gemini if you want it — Google AI Studio and Vertex adapters both ship. Or Claude, or GPT, or Mistral, or a local model. Different assistants can use different models, and switching is a setting rather than a change of vendor.
The same assistant reads Google Workspace, Microsoft 365, Nextcloud and your ITSM. Work does not usually sit in one suite, and an assistant confined to one of them can only ever see part of the picture.
Install the detector and personal data from a document or a thread is tokenised on your own hardware before any prompt leaves for a provider, and restored in the answer. When the model is on your own network, nothing leaves at all. That layer sits outside the suite, which is exactly why it can apply to every suite you connect.
Gmail, Calendar, Drive, Docs, Sheets, Slides, Contacts, Groups, Keep and Maps — ten Google services, and Meet recordings are ingested into meeting notes with transcription and speaker identification. Each person connects their own account by OAuth and an integration acts as the person who authorised it, so an assistant can never read a mailbox its user could not open themselves. This is our most complete connector set, not an afterthought.
When the suite, the model and the workspace are one vendor, every question about processing has the same answer and you cannot change one part without changing all of it. Splitting them means you can keep Google Workspace as your office suite and still decide, separately, which model sees which data and in which country — including deciding that some categories never leave your own network. That separation is the product.
Trying it
OAuth on your own credentials, nothing org-wide by default. Ask an assistant something about your own Drive and your own inbox and judge the answer against what you would get in the sidebar.
Gmail · Drive · Docs · Sheets · Calendar · Meet recordings
Change the assistant from Gemini to Claude, or to a model on your own network, and run the same question. Nothing about the assistant, its knowledge or its tools has to be rebuilt — that portability is the thing a single-vendor suite cannot offer.
Gemini → Claude → local Qwen · same assistant
With the detector installed, watch what the provider actually receives. That is the comparison in one step: not which model writes better, but what left your organisation to get the answer.
tokenise on the way out · restore in the reply
Honest boundaries
Questions
No, and most teams do not. Google Workspace is the deepest integration Bee Flow ships — ten services — so it sits beside your suite rather than replacing it. The suite stays your office; Bee Flow becomes the layer where AI, automations and knowledge live.
Yes. Both the Google AI Studio and the Google Vertex adapters ship, so you can keep Gemini and change only where the workspace runs. That is a common configuration and a completely reasonable one.
No. Bee Flow talks to those apps through their APIs — reading a thread, filing a document, updating a sheet — but there is no in-document sidebar. If your main use is drafting inside the document you already have open, that is a real gap and you should weigh it.
They are ingested into meeting notes: transcription with speaker diarisation, and per-person voiceprints for identification where you have enrolled them. That runs on your own WhisperX service when you self-host, so the audio does not have to leave your infrastructure.
Point the assistants that touch it at a local model — Ollama, vLLM or any OpenAI-compatible endpoint on your own network — and no prompt from those assistants reaches a cloud provider. Local models are recognised and the detector skips them; anything unknown is treated as external and inspected.
Ten Google services, your own documents, and a model you pick. Fifteen minutes will tell you whether the separation is worth it for your organisation.
See the integrations How sovereignty works