Notion integration
Your assistant reads and writes in the workspace you already keep. It finds the page, reads what is on it, creates the task, moves the due date, and leaves the comment — on exactly the pages you chose to share and nothing else.
Connecting Notion
Open the assistant and choose Notion
The integrations list is in the assistant's tools section, alongside the other workspace tools.
Authorise, and choose what to share
You are redirected to Notion. This is the important screen — you pick the pages and databases the connection may reach. Share the project database, not the whole workspace.
Decide reading from writing
Search, page properties, page content and database structure are the read tools. Create page, update page and add comment are the write ones. A read-only assistant is a genuinely useful thing on its own.
Turn on the database tool if it will write
Creating and updating rows works far better when the assistant can first read which statuses and tags your database actually accepts. Leave it off and expect rejected values.
Try both halves in the Playground
Ask what the status of something is, then ask it to create a task in a named database. Open Notion and check the properties it filled in.
What a workspace costs once it is big
The answer is written down and nobody can find it
The decision was made, the brief exists, the spec is somewhere. Notion search finds forty pages with that word in them and you read six before giving up.
Capturing a task breaks the thing you were doing
Something needs doing on Friday. Writing it down means leaving the conversation, finding the right database, filling the right properties. So it does not get written down.
The status is right in the database and wrong in everyone's head
Half the team asks each other what state something is in, because opening the board is three clicks more than asking the person next to you.
Most tools want the whole workspace
An integration that reads everything is one leak away from being a serious problem. The permission you want is per-page, and almost nothing offers it.
Seven tools, and Notion decides what they can reach
The access model is the part to understand first, because it is unusually good. During authorisation you pick which pages and databases the connection may see. The assistant can act on those and on nothing else — not the rest of the workspace, not pages shared with you later by someone else. To widen or narrow it afterwards you change the connection's page access from inside Notion, not from Mingleego. That means the boundary is enforced by Notion and visible to you in Notion. Inside that boundary there are seven tools. Four read: search the workspace by name, read a page's tracked properties, read the text written on a page, and read a database's structure with its valid status and tag options. Three write: create a page, update one, and add a comment. One split catches people out, so it is worth stating: reading a page's properties and reading a page's text are two different tools. Properties are status, due date, assignee — the fields the database tracks. The text written on the page is the other one, and long pages are read in chunks automatically. An assistant that should answer "what is the status of the launch" needs properties; one that should summarise the meeting notes needs content. The database tool is the quiet one that makes writing reliable: before creating or updating a row, the assistant reads what the valid statuses and tags actually are, instead of inventing a value your database will reject.
What it is actually good at
Capturing the task in the conversation
"Remind me to call the client on Friday" becomes a row in the right database with the right properties, without leaving the chat.
Answering from what is written down
What did we decide about pricing. The assistant finds the page, reads it and answers — instead of returning a list of pages for you to read.
Status without opening the board
Where is the launch, who owns it, when is it due. Properties, read live, phrased as a sentence.
Closing things out
Mark it done, push the date a week, add a tag. Small updates that are only annoying because of how many clicks they take.
From whatever channel your team uses
Telegram, Slack, Discord, the web widget — the same workspace, the same shared pages, the same assistant.
Summarising the long page
Long documents are read in chunks, so "what is in the meeting notes" works on a page nobody has time to read in full.
The seven tools
Search the workspace
Finds a page or database by name — how the assistant locates anything it has no direct link to. It searches only what was shared with the connection, so the results are already scoped.
Read page properties
Status, due date, assignee and the rest of the tracked fields. This is what answers "where is that" and "who has it" — and it deliberately does not include the text on the page.
Read page content
The actual text: meeting notes, briefs, long documents. Long pages are read in chunks automatically, so the length of the page is not your problem.
Read a database
Its structure and the status and tag options it accepts. Used before writing, which is the difference between a row that lands correctly and one your database rejects.
Create and update pages
A task, a note, a meeting, a project — created with the details given, and updated later to mark it done, push a date or add a tag, leaving everything else untouched.
Add a comment
Leaves context on a page where the team already looks for it, rather than in a chat transcript nobody will open again.
Before you connect a workspace
Only what you shared with the connection when you authorised it. Everything else in the workspace is invisible to it, including pages someone shares with you afterwards. That boundary is enforced by Notion, which is why it is trustworthy.
Share one database and see what happens
Connect Notion, share a single project database, and give the assistant only the reading tools. It is the cheapest way to find out whether your team would use this.
Create an assistant