I built a Notion-style workspace my AI agent keeps current (prompt included)

What is a workspace your agent keeps current?

A workspace, in the Notion or Coda sense, is a small web of pages: a home page at the top, pages nested under it for projects and areas, and structured tables (linked databases) for the things you track as records rather than prose. It is where your notes, plans and lists live together instead of being scattered across separate documents.

The version built on Pages looks the same as one you would recognise. The difference is who does the upkeep. You write to the workspace by talking to your AI agent, and it reads from the workspace the same way. So instead of opening a page, finding the right spot, and typing, you say "mark the onboarding task done and add a note that the client signed off," and your agent finds the row, flips the status, and writes the note on the right page. The structure is yours. The filing is the agent's job.

What can you put in it (nested pages and linked databases)?

Two kinds of thing, and a workspace is mostly a mix of both.

Nested pages are for prose and outline: a home page that links out to everything, a page per project, an area page for "Finances" or "Hiring," a running notes page. Pages can sit inside pages, so the structure mirrors how you think about your work rather than forcing everything into one flat list.

Linked databases are for the parts that are really records: a tasks table, a contacts list, a table of content ideas. These are backed by a Database, a real structured table underneath, so each row has fields you can filter and sort, and your agent can read it as data rather than scanning paragraphs. A task row has a status, an owner and a due date. A page can reference that table, so your project page shows the tasks that belong to it.

Put together, that is a workspace: pages for the narrative, databases for the records, and links between them. A project page that opens with what the project is and then lists its open tasks from a shared database is the standard shape, and it is the one the prompt at the end sets up.

Who is this for?

People who already think in workspaces. If you keep a personal or team wiki, run projects out of nested pages, or have a tasks table you wish updated itself, this is built for you. It is built for personal AI agents, so the assumption is that you already work through Claude or ChatGPT and want them to maintain real structure rather than answer in a chat window that forgets everything next time.

It is a good fit if the upkeep is what puts you off. Workspaces rot when the filing is manual, because there is always something more urgent than tidying a page. If you hand that maintenance to your agent, the workspace stays current as a side effect of you telling it what happened.

It is a weaker fit if you need rich visual layouts, real-time multiplayer editing, or a heavy app to live in all day. AgentDocs is the place your agent reads and writes; it is not trying to be the screen you stare at for eight hours.

How is this different from a workspace you maintain by hand?

The structure is identical. Nested pages, linked databases, the same building blocks you would set up in a Notion or Coda workspace. We use those names because that is the category and that is how people search for it, not because we are claiming to replace what you already use.

The one real difference is upkeep. In a workspace you maintain yourself, every change is a manual edit: open the page, find the line, type, save. That work is small each time and large in total, and it is the first thing to slip when you are busy. Here, the agent does that part. You report the change in plain language and it makes the edit in the right place, whether that is a status on a task row, a new bullet on a project page, or a fresh nested page for a project that has only started.

The second difference is that the workspace reads back. Because your agent both writes and reads it, you can ask across the whole thing: "what is open across all my projects," "which contacts have I not followed up," "summarise what changed this week." You get an answer drawn from the live workspace instead of clicking through pages to assemble it yourself. That is the same upkeep principle behind the linked-database CRM and the content calendar: the structure is easy to set up, and letting your agent maintain it is what keeps it worth having.

Because it is built on Stash with the same Google sign-in as the rest of your account, the workspace is not tied to one assistant. Ask Claude to update it today and ChatGPT next week, and both read and write the same pages. On cost: AgentDocs has a free tier with no per-seat fees, you connect over one link with no install and no card, and there is a Pro tier for heavier use. It is a new product, so pricing may change as it settles.

How do you set one up?

You describe it once and your agent builds it. You do not create pages by hand, pick a template, or wire up the database fields yourself. You tell your agent the shape you want (a home page, these nested pages, a tasks database linked to the project pages) and the rules for keeping it current, and it does the construction and the upkeep from there.

Works in Claude.ai, Claude Code (fastest) and ChatGPT, all over MCP. You connect once at https://getagentdocs.com/mcp, with no install and no card, then paste the prompt below.

The exact prompt

Paste this to your AI agent once AgentDocs is connected. Edit the brackets to match the projects and areas you have:

Set up a personal workspace for me in AgentDocs on Pages.

Build this structure:
- A home page called "Workspace" with a short intro line and links to
  everything below.
- Nested under it, one page per project I name: [Project A], [Project B].
  Each project page starts with a one-line summary of what the project is.
- An "Areas" page with a nested page for each ongoing area:
  [e.g. Finances, Hiring, Reading].
- A linked Database called "Tasks" with these fields: Title, Status
  (Todo / Doing / Done), Project, Owner, Due date.

Link the Tasks database to the project pages, so each project page shows
the tasks whose Project field matches it.

Then keep it current for me by these rules:
- When I tell you something changed, edit the right page or task row.
  Don't make me say where it lives; find it.
- When I mention a new task, add a row to Tasks with a Project and, if I
  gave one, an Owner and Due date. Default Status to Todo.
- When a task is finished, set its Status to Done rather than deleting it.
- When I start a new project, add a project page nested under Workspace and
  link its tasks.
- Don't invent statuses, owners or dates I didn't give you. If something is
  unclear, leave it blank and tell me.

Build the structure now, confirm what you created, and ask me for my first
few tasks so you can fill in the Tasks database.

That is the whole setup. No template to pick, no fields to wire up. You describe the workspace once, your agent builds it on Pages, and from then on you tell it what changed and it keeps the workspace current.

This is the fifth in our "Built in AgentDocs" series, real things built by talking to an AI agent, with the prompt included every time. Next up: turning one of those pages into a slide deck your agent drafts for you, built on Slides.