How can we help?

Profusia is a place for AI-generated documents to live. You connect your AI assistant (Claude, ChatGPT, or anything else that speaks MCP) to Profusia, and it can publish files straight into your workspace — web pages, dashboards, Markdown, code, images, PDFs, Office documents. Published documents get a private URL you can share, they keep a full version history, and they can be arranged into a site with navigation. Pages can be kept up to date automatically with live data. The short version: your AI makes something, and instead of a file in a chat window you get a real link that stays current.

Profusia connects to Claude as a custom connector. In Claude, go to Settings → Connectors → Add custom connector, and give it your workspace's MCP URL, which is your Profusia address followed by /mcp — for example https://yourteam.profusia.com/mcp. Claude will walk you through signing in (Profusia uses standard OAuth). Once it is connected, just ask Claude to publish something: "publish this as a page in Profusia" works. There is also a static access-key option if you prefer, which your workspace owner can issue.

ChatGPT works too, and there is nothing extra to set up on Profusia's side — the connection uses the same open standard (MCP). In ChatGPT, enable developer mode and add a custom connector pointing at your Profusia MCP URL (your address followed by /mcp). When it asks for authentication, the API-key option takes your workspace access key. Everything works the same as it does in Claude: publishing, updating, share links, live data. ChatGPT's scheduled tasks can also refresh live data on a schedule, the same way Claude's Routines can.

Ask your assistant to publish. For example: "publish this report to Profusia" or "put this dashboard in Profusia under the Client Work folder". Your assistant will write the file and send it over. You get back a link, and Profusia tells the assistant where the file landed so it can tell you the folder too. You can publish HTML pages, JSX (which Profusia compiles in the browser — no build step), Markdown, code files, images, PDFs, and Office documents. Anything Profusia can render becomes a live page; anything it can't becomes a clean download.

If the file already exists — on your computer or phone, exported from another app, attached to a chat — don't make your assistant retype it. Ask for an upload link. Your assistant can call get_upload_link, which gives you a private drag-and-drop page. Drop the file in and it lands in your workspace like anything else. That works for any file type, and it is much faster than streaming a large file through a conversation.

Every document is private by default. Nobody can see it until you create a share link. Ask your assistant for a share link (get_share_link) and you get a URL with a secret token in it. Anyone with that link can view the document — treat it like a key. Share links can be set to expire, and they can be revoked and reissued at any time. If you want to look at your own document, you don't need a share link at all: sign in and open it from your site. Signing in shows you everything you have access to. There is also a sharing review (review_sharing) that lists every live link in your workspace, so you can see at a glance what is currently reachable from outside.

These do different jobs, and it is worth getting the distinction: A FOLDER is where a file lives — the filing cabinet. One file, one folder. A PROJECT (some workspaces call them Initiatives, Campaigns, or something else — you choose the word) is what a file is FOR. Projects can nest, a file can be in one, and each project carries a description and a set of goals. The reason projects matter: those goals get read back to your assistant whenever it publishes into that project or lists what is in it. So a session next month picks up what the project is trying to achieve without you re-explaining it.

A site is how your documents are presented to a reader — a navigable place rather than a pile of links. A site is made of SPACES, which nest up to six levels deep. Each space gets an overview page generated from what is actually in it: real counts, what changed recently, cards for the spaces inside it, and live previews of the documents. You can build a whole site in one call from a template that mirrors your existing folders, then refine it — either by asking your assistant, or by hand in the visual editor. Sites are live. There is no publish button. What controls who can see a site is simply whether a link to it exists.

Each space picks a presentation shell — how it displays, as opposed to what is in it. The choices are Showcase, Repository, Rail, Narrative, Handbook and Shelf. Showcase is the default at the top of a site; Repository is the default inside it. A space inherits its shell from its parent unless you set one. You can change it by asking your assistant (set_space_shell), or use the View picker on the space itself. You can also preview a different shell without committing by adding ?shell=rail (or any other name) to the URL.

Every document has a comment thread behind a small toggle, and every space has its own discussion for talking about the place rather than a single file. A comment on a document can be PINNED to a specific passage or image: select the text, and your comment points at exactly that. The highlight is drawn over the document — the document itself is never modified. Comments are for signed-in members of the workspace, any role. Someone holding a share link is anonymous, so they see the document but not the discussion. The useful part: your assistant can read the comments (read_comments) and revise the document based on them, then post a note saying what it changed. Pinned comments come through with the passage they point at, so "this number is wrong" is actionable. If you are hunting for a comment you remember, there is a Comment Centre that shows every comment across the workspace, filterable.

Every time a document is updated, Profusia keeps the previous version. Nothing is overwritten. To go back, ask your assistant to roll the file back (rollback_file) to an earlier version. The share link never changes, so anyone holding it just sees the restored content.

Deleting is reversible. When you delete a file, folder, site, project or dataset it goes to the workspace trash — it stops appearing in listings, and its share links stop working immediately, but it is not gone. You can see what is in the trash (list_trash) and restore it (restore_item). If you delete a folder, restoring it brings back what was inside as a group. Permanently erasing something is a separate, deliberate step (purge_trash) that you have to confirm. Nothing is ever swept away automatically. Renaming and moving work on anything: files, folders, sites, spaces, projects.

A document can be backed by a small dataset so its numbers stay fresh without republishing the whole page. The flow: your assistant creates a dataset (create_dataset), binds it to the page (bind_file), and from then on updating the data (update_dataset) updates what the page shows. The best part is scheduling it. Claude can set up a Routine — or ChatGPT a scheduled task — that fetches whatever the numbers come from and calls update_dataset on a schedule. Your dashboard then updates itself every morning without anyone opening it. The Live data view on your site lists every dataset and what it is keeping current.

You can embed a Profusia document into SharePoint, Confluence, Notion, Teams, or an intranet page. Ask your assistant for embed code (get_embed_code). It will ask where you are putting it rather than asking for technical details, and gives you something you can paste. The embedded view shows the document plus a small strip saying what it is, how current it is, and how to get back to the full version. Pasting a plain link into Slack, Teams or Notion also produces a proper preview automatically. Embed access is a separate permission from a share link: forwarding a share link does not let someone embed the document. You can revoke an embed at any time (revoke_embed).

Every document has "Open in Claude" and "Open in ChatGPT" buttons, plus a Copy button that gives you a prompt about the document you can paste into any AI tool. The app the document was originally published from is listed first, and if that app has a connector to your workspace the prompt carries the document's id, so your assistant can pull the actual document rather than guess from a description.

Profusia can record the moments a project's direction moved — a branch, a pivot, something abandoned — with the reason and links to what it replaced or revised. Your assistant records these as they happen (record_turn), because it is in the conversation where the decision gets made. Publishing a document is not one of these moments; a real change of direction is. Later, get_course reads the whole thing back: what the project is doing now, what was tried and dropped, and DRIFT — the documents last written before the direction changed, which may still assume the old plan. If a project has no recorded history, Profusia says so rather than inventing one from publish dates.

A workspace has four roles: owner, admin, member and viewer. Owners and admins can invite people and manage the workspace; members can create and edit; viewers can read and comment. To invite someone, ask your assistant (invite_member) or use the admin area. Registration is invitation-only — you cannot sign yourself up to somebody else's workspace. You can also create groups and use them to narrow who can see a particular document (set_file_audience), which is useful when most of a site is for everyone but a few documents are not.

The main points: Documents are private by default and only reachable with a link you created. Every workspace is served from its own subdomain, so one workspace's content cannot be loaded from another's address even with a valid link. Hosted pages run under a strict security policy: a page can only fetch its own dataset, and Profusia never executes anything from a document on the server. Profusia stores no third-party credentials. It has no access to your email, your Google Drive, or your Jira, and it never asks for them. All tokens and keys are stored hashed, expire where appropriate, and can be revoked. There is a fuller public write-up at /trust, which is the page to send to a security reviewer.

There is a short guide at /skill written for your AI rather than for you. Saving it into Claude or ChatGPT (as a skill, project instruction, or custom instruction) makes your assistant noticeably better at using Profusia — it stops asking you where things go and starts recording project history properly. It is a public page, so you can hand the link to a colleague without giving them access to anything.

Being straight about what Profusia does not do today: There is no published pricing and no self-serve signup — Profusia is in development and workspaces are set up by hand. Email is not wired up yet, so invitations are links you pass on yourself, and there is no "forgot password" email. A workspace owner resets a password. Single sign-on is built but switched off, pending configuration. Profusia does not connect OUT to other systems — it will not read your Jira, Salesforce, or Google Drive, and this is a deliberate design decision rather than a missing feature. Your AI assistant already has those connections; the intended pattern is that your assistant reads them and publishes the result into Profusia. There is no data export tool yet. It is on the list and considered important.