AI in Flowra: where, with which plan, and what it does with your data
AI in Flowra does a few specific jobs: meeting transcripts and summaries, writing and asking in Knowledge, the page assistants and the chat assistant. This page says where each one is, which plan turns it on, what its limits are, which language it answers in, what data goes where and how it switches off.
Open in Flowra/knowledgeAI
MeetingTimestamped text, summary and tasksKnowledgeAsk Flowra, writing assistantAssistantsWith AI or without itPlanPrime, per personGatesAdmin switch, plan, access, serviceTranscription does not know who spoke, and invents no name.
What this is wired to
Where AI is in Flowra
In a meeting, the “Meeting AI assistant” makes the meeting's transcript from its recording and, from the transcript, writes the summary, decisions and tasks. In Knowledge, “Ask Flowra” answers from the team's pages, and the writing assistant inside each page summarises, rewrites or translates text. The page assistants on Actions, Knowledge, Folio, Calendar, Projects and Companies, and the pinned chat assistant, draw on it too.
Some things that might look smart have no model behind them: people suggestions in My network are worked out with clear rules, and when no model is available the assistants move ahead with ready-made questions and templates and carry the “Without AI” label.
Meetings: text line by line
On the page of a finished meeting, the host or any participant presses “Transcribe the recording”, or, if there's no recording, “Upload audio and transcribe”. The file has to be your own, audio or video, and up to 25 MB. The result is a list of lines, each with its time; tap a line and the recording plays from that moment.
Transcription doesn't know who spoke. Making up a name for every line would be a confident lie, so each line gets either the name of someone Flowra is sure about or the neutral label “Speaker”.
- Open the page of a finished meeting
- Press “Transcribe the recording” or “Upload audio and transcribe”
- Tap any line to play the recording from there
Summary, decisions and tasks
“Summary and action items” writes a summary paragraph, key points, decisions and tasks from the meeting's transcript, in the language the meeting was held in. With the “Turn the tasks into actions” tick, each task becomes a real action with its own history; pressing the button again doesn't double the list, and an owner is set only when the name spoken matches one of the participants, because a wrong owner is worse than no owner.
If the model's answer can't be read, nothing changes and the error is reported; an empty summary never overwrites a good one. Under the summary it says when and with which model it was made, and “Save to Knowledge” turns it into a Knowledge page.
Ask Flowra
On the Knowledge home, the “Ask Flowra...” box answers only from pages you yourself can read. The question's words are weighed against the pages' titles and text, a few of the best pages go to the model, and the sources are shown numbered beside the answer. If nothing is found, it says so and makes nothing up.
The writing assistant in Knowledge pages
Inside every page, the “AI assistant” works on the selected part or the whole page: summarise, improve the writing, rewrite, shorter and longer, change the tone, translate, pull out actions and decisions, suggest a title, build headings, continue the text and explain simply; you can also ask questions about that page.
Nothing is written on the page without your own choice: “Add to page”, “Replace selection”, “Copy” or “Again”. Tick the extracted actions and create them with “Create * actions”. Under every answer it reminds you that an AI answer can be wrong.
Which plan
The AI features are part of the Prime plan: meeting transcripts and summaries, Ask Flowra and the writing assistant, and the AI answers of the page assistants. Free and Plus have the rest of Flowra. The plan counts per person, not per workspace: the best plan paid for you, whether your own or that of the owner of a workspace you belong to; a meeting has no workspace at all, only a host and participants.
Without Prime, every button says why it won't work before you click, and shows the “Upgrade to Prime” card. The check happens on the server, not just in the buttons. The chat assistant is the exception: it has smart answers on every plan, just with a smaller daily allowance.
Usage and limits
The chat assistant has a hundred smart answers a day on Prime and ten a day on the other plans; the count left for today is shown under its suggestions, and after that the answer comes from the Flowra guide. A meeting audio file up to 25 MB is transcribed. Each page assistant message carries up to eight attachments, each file up to 5 MB for reading, and long text is shortened before it goes to the model; a very long meeting transcript is also shortened from the middle for the summary, so the opening and the final decisions stay.
Every call, successful or not, is recorded: the kind of job, the model, input and output tokens and seconds of audio. The totals show in the admin panel's overview, because a feature that spends money should be a number the operator can read.
Which language it answers in
Flowra sends the language you work in with every request, and the answers of the assistants, Knowledge and the chat assistant come back in that language, in any of Flowra's nine languages. A meeting summary is written in the language the meeting was held in, a translation in Knowledge in the language you choose, and a Folio assistant design in the language you gave for the sheet. Change the language in Settings and the next answers come in the new one.
Where your data goes
For each answer, only what that job needs is sent to the AI service: the meeting transcript for a summary, your own audio file for transcription, pages you can read for “Ask Flowra”, the selected text for the writing assistant, and the request, answers, attachments and needed context for the assistants. A file you can't open isn't read, and a page you can't read never reaches the model. The meeting panel says exactly this next to its buttons: the meeting's text is sent to OpenAI.
The service today is OpenAI, and the operator sets its address and model on the server. The usage log keeps the text of no request or answer; what stays is what you've made yourself, such as a meeting's transcript and summary, a page or an action.
Turning it on and off
In the admin panel, under Features, the operator turns “AI” off or on for everyone, and the server enforces it too. When it's off, the meeting panel says the feature has been turned off, the assistants work with ready-made questions and templates, and the chat assistant answers from the guide. If the service key isn't on the server, the buttons say so.
You have choices too: on a plan without AI, press “Continue without AI”; in a meeting, untick “Turn the tasks into actions”; and edit or drop any draft before you confirm.
Four gates, cheapest first
Every AI feature is checked in the same order: the operator's “AI” switch, then the plan, then your access to that meeting, page or project, and last of all the outside service, the only costly gate. In the assistants, the first two gates lead to the path without AI, and the access gate always says no.
Every service failure has its own message: wrong key, quota used up, rate limit, model unavailable, timeout, or a region where the service doesn't work.
Your own AI, with MCP
If you have your own Claude, ChatGPT or Cursor, connect it from the MCP page under Connectors. On this route Flowra runs no model: your own model does the thinking, Flowra only hands it the tools with the permissions you choose, and everything it does is in your name and under the same app rules. That's why MCP doesn't need Prime and only has its own rate limit. The details are in the MCP guide.