# An AI workspace where agents work on the real documents.

Ask Studio AI beside any document or on the map, or connect Claude, ChatGPT, Cursor, Claude Code or VS Code over MCP. Both read the whole workspace, write block by block and stay inside the role of the person they work for, and Studio AI cites the passage behind each statement. Every write keeps the version before it.

Updated 23 October 2026

![A project on Nodary's map: 43 documents grouped around eight folders, linked by hierarchy lines and labelled links, with teammates' cursors on the map.](https://nodary.ai/marketing/map-constellation.webp)

_Saltmarrow is a sample project: its studio, team and game are fictional._

## Studio AI, beside the page you are on

Studio AI reads the whole workspace and works where you are. It writes and edits documents block by block, and files, links, tags, dates and comments for you. Chat answers questions, Agent carries a long task through, and Archiver reads a project and writes knowledge notes that later sessions start from. It looks at the pictures, moodboards and boards in your documents too. Select any words and it clarifies them, shortens them or changes their tone. When you do not know where to begin, start a context from a mission: six for projects, four for personal work such as travel or studies.

- Three modes: Chat, Agent and Archiver
- Rewrite a selection, then replace it or insert below, with Undo
- Each answer says what it did and what it cost in Comets
- A commenter's Studio AI comments; a reader's only reads

What it does not do: Studio AI spends Comets: on Free, 20 a month per person, shared by the free workspaces they own; on Pro and Studio, the team's pool of 800 or 1,500 per seat, with caps per member.

![A document open in Nodary's editor with its outline and details on the left, the page in the middle and the document's AI assistant on the right.](https://nodary.ai/marketing/document-pitch.webp)

## Your own agents, over MCP, on every plan

Connect Claude, ChatGPT, Cursor, Claude Code or VS Code to your workspace. An agent signs in with OAuth, through a consent screen, or with an API key that can be kept to one project, and one connection can work in several workspaces you tick. 98 tools reach every part of the product: documents block by block, sheets cell by cell, boards seen as pictures, pictures with their drawn notes, properties, saved views and versions. Each person connects their own agent, with their own role, whether they work alone or in a large team.

- Included on every plan, Free within its limits
- 120 requests a minute per connection, and no daily write quota
- 25 documents created or 50 changed in one call; a retry never writes twice
- Files in and links out, so a render or a 3D model travels between tools

What it does not do: Share links, document access, members, roles, keys, billing and deleting a project stay human decisions: no agent can make them.

![The API keys and MCP section of a Nodary profile: the MCP address to paste into Claude Desktop, Claude CLI or any MCP client, and the owner's agent key for the workspace.](https://nodary.ai/marketing/mcp-connect.webp)

## Agents know what is true before they answer

Every document can be canon, exploration, obsolete or superseded by another, and an agent is told before it reads the words. Studio AI cites the part of the document behind each statement, and shows what it read then against what the page says now. When two documents disagree, a consistency scan reports both passages side by side, quoted word for word; a quote that is not in the document is refused, and a resolution is only proposed when you ask. For any other AI tool, a context brief gathers up to 12 sources, shows its size, says which sources changed since you copied it, and exports as Markdown.

- Truth status given to agents before the text
- Answers that point to the passage they rest on
- Contradictions shown as two passages, decided by you
- Context briefs of up to 12 sources, taken anywhere as Markdown

What it does not do: A context brief is a copy: it does not update itself, it tells you which of its sources changed so you can prepare it again.

![A document opened after a teammate changed it: a note says what changed since the last visit and by whom, and the version history compares that version with now.](https://nodary.ai/marketing/what-changed.webp)

## Agents work with your rights, and cannot erase your history

An agent only does what the role of its person allows, on the server, for every call. Every agent write keeps the version before it, and agents' versions never push out people's, so any change can be compared and put back. An agent's comments say they come from an agent, and each change it makes is signed in your name in the activity feed. A retry with the same key never writes twice, so a dropped connection does not leave two copies behind.

- Roles enforced on every call, for people and agents alike
- A version kept before every agent write
- 50 versions kept for people and 50 for agents, apart

What it does not do: A document an agent deletes goes to the bin like any other: it comes back for 30 days, then it is gone.

![A workspace's activity feed on Nodary's home: teammates commenting, deciding pictures, requesting and approving reviews, changing statuses and editing documents, newest first.](https://nodary.ai/marketing/activity-feed.webp)

## Who it is for

- [MCP knowledge base](https://nodary.ai/solutions/mcp-for-project-knowledge): A knowledge base your agents read and write, with your permissions.
- [AI context management](https://nodary.ai/solutions/ai-context-management): The context an AI conversation starts from, and a way to know when it is stale.
- [Product and engineering teams](https://nodary.ai/solutions/product-and-engineering-teams): Requirements, technical designs and decision records linked to the work they explain.

## Ask for the whole job, not one edit

Over MCP, an agent reads a project's outline with get_tree, finds passages with search_documents, writes with write_document or edit_document, and records what disagrees with report_contradiction, quoting both documents. save_context_brief keeps a set of sources an agent can read back with get_context_brief, and get_canon checks a draft against what the project holds true. get_document warns an agent when a document is superseded, and names its replacement.

> Read the launch project, find any two documents that disagree on the release date and report them, then write a one-page summary of what is decided, citing each source.

Tools it calls: `get_tree`, `search_documents`, `report_contradiction`, `write_document`

## Start from a template

- [PRD template](https://nodary.ai/templates/prd): A product requirements document an agent can read, check and keep current.
- [Technical design document template](https://nodary.ai/templates/technical-design-document): The design of a system, its choices and its open questions, ready for review.

## Questions

### Which AI agents can connect to Nodary?

Any client that speaks MCP over HTTP. Claude, ChatGPT, Cursor, Claude Code and VS Code have setup steps in the MCP documentation. Agents sign in with OAuth or an API key and work within the role of the person who connected them, and one connection can work in several workspaces you tick.

### Is the MCP server paid?

No. It is included on every plan, Free included within the Free plan's limits, with up to 120 requests a minute per connection and no daily write quota. Agents you connect run on their own model and spend no Comets. Every client sees the same 98 tools, listed by job in the MCP documentation.

### What does Studio AI cost?

It spends Comets. On Free, each person has 20 a month, shared by the free workspaces they own; on Pro and Studio, the team shares a pool of 800 or 1,500 per seat, with caps per member, and the 14-day trial includes 200 for the workspace. Each answer says what it cost, and on paid plans Comet packs add more when the pool runs low.

### Does our content train the AI models?

No. Studio AI runs on Claude models through Amazon Bedrock, and your content is not used to train them. AI requests are processed in the United States.

### Can an agent overwrite or delete our work?

Only within the role of its person, and never without a way back: every agent write keeps the version before it, agents' versions never push out people's, and a deleted document comes back for 30 days. Access, members, roles, keys, billing and deleting a project stay with people, and each change an agent makes is signed in the name of the person it works for.

### Can I use my own model?

Yes, through MCP: any MCP client with its own model can read and write your workspace within your permissions. Studio AI, inside the app, runs on Claude models through Amazon Bedrock. You can use both: Studio AI beside the page, and your own agent in an editor or a terminal.

## Read next

- [Knowledge base](https://nodary.ai/knowledge-base): A team knowledge base whose pages link to each other, say which one is the truth, and keep a history to restore, for people and AI agents alike.
- [Reviews and approvals](https://nodary.ai/reviews-and-approvals): Review requests, approvals with a note, comments on the exact passage or point of a picture, and every decision signed with who and when.
- [Sheets and databases](https://nodary.ai/sheets-and-databases): A real spreadsheet inside any document, charts drawn from it, and typed properties with live views that turn documents into a database.
- [Glossary](https://nodary.ai/glossary): Plain definitions of the words used on this page, from MCP to context brief.

## Give your agents the same documents your team trusts.

Start free, connect an agent in a few minutes, and keep every change it makes within reach. Free for two members and ten documents. No card.

The page: https://nodary.ai/ai-workspace