Comparison

Busabase vs ChatGPT Space

ChatGPT Space, which OpenAI launched on 29 September 2026, is a shared home for a team’s pages, files, spreadsheets, presentations, and Sites, where people, ChatGPT, Codex, and Dots work on the same material in real time. Busabase is an open-source database and workspace for AI agents — agents and people share the same structured data, docs, skills, and apps, and writes that matter become reviewed, trusted records. The two overlap less than the word “workspace” suggests. Space is built around documents your team co-edits with OpenAI’s agents. Busabase is built around records that any agent can write, where a consequential change can wait as a reviewable proposal before it becomes true.

This page is for teams deciding whether Space also covers the data their agents produce, or only the documents. It says plainly where Space is the better product — for a team that already works in ChatGPT, it often is. It also separates what OpenAI documents, what launch coverage reported, and what is our own inference.

Last verified: 30 September 2026, one day after launch, against OpenAI’s help documentation and launch coverage. Anything we could not find in those sources is marked “not described”, which is not the same as “cannot”.

The short verdict

What ChatGPT Space ships today

A comparison written from a first impression is worse than no comparison, so here is what was published, and by whom.

From OpenAI’s documentation. Space is “a home for your pages, files, and shared work with ChatGPT.” It holds Pages (editable documents that can contain tables), spreadsheets, presentations, Sites (websites and tools built with ChatGPT), images, and uploaded files, grouped into spaces by topic or team. Share a Page with edit access “to write together in real time.” Text ChatGPT generates can be reviewed in place: you can “accept it, ask for a change, or undo it.” Comments can mention ChatGPT, and attribution shows who contributed where available. Connected apps such as Google Drive keep their original permissions, and sharing a Page does not expose private chats or saved memory.

For Enterprise, admins manage sharing at workspace and role level and can switch off new sharing. Saved Library files follow the workspace retention policy, and the Compliance API can export or delete eligible files. Space replaces Library on Pro, Business, and Enterprise plans, on web and desktop first, with mobile to follow.

From launch coverage. VentureBeat reports that employees, ChatGPT, Codex, and Dots — OpenAI’s always-on agent coworkers, each with its own cloud computer and browser — work against the same shared context, with several people each using their own ChatGPT on the same Page at once. It also describes an auto-review system that “checks potentially consequential actions against the user’s instructions, Custom Rules and OpenAI’s built-in safety requirements” before deciding whether work proceeds on its own or needs approval. TechCrunch notes that Pages can carry standing instructions and frames the launch as OpenAI taking on Microsoft’s workplace suite; it reported slides still rolling out. Pricing for Dots has not been disclosed.

That is a serious product on day one. “ChatGPT is a chat window, not a workspace” stopped being true this week.

The question Space does not ask

Space is designed around a real question: how do a team and its AI write the same document without stepping on each other? Accept-or-undo in place, attribution, comments, and an auto-review of consequential actions are good answers to it. The question that decides this comparison sits one step further along:

When an agent changes a value that other agents, reports, and apps will read as fact, who decides it becomes true — and where is that decision kept?

Accepting or undoing a paragraph is a review of prose, by the person looking at it, at the moment it is written. It works because the reader and the reviewer are the same person. A field in a record is different. The next reader is usually not a person, and the review has to survive as part of the record: what the value was, what it became, who proposed it, who accepted it.

Auto-review answers a neighbouring question — should this agent take this action now? That protects the action. It is not the same as deciding whether the resulting value belongs in canonical data. As of 30 September 2026, OpenAI has not described a schema’d record store, field-level diffs or proposals for agent writes, version history, or a public API or MCP route into Space for agents outside OpenAI’s own.

Why the difference is structural

This is not a gap we expect a point release to close, because it follows from what each product is for.

Space’s unit is the document, and its agents are OpenAI’s: ChatGPT, Codex, and Dots, working inside OpenAI’s product with your team’s context. That is exactly what makes the co-editing experience coherent — one vendor owns the editor, the model, and the agent.

Busabase’s unit is the record in a Base with a schema, alongside docs, skills, and apps, and its agents are whichever ones you already use: Claude Code, Cursor, Codex, OpenClaw, or ChatGPT itself over MCP. When agents from three vendors write into the same data, the review cannot live inside any one of their editors. It has to sit on the write path of the data and stay attached to the record afterwards.

So “can AI work on our team’s material?” gets a yes from both. The dividing line is whether that material is a document people read or a record that systems read.

Comparison by the job you are trying to do

Read the rows, not the totals. If your real job is the first three rows, Space is the natural choice and you can stop here.

Where ChatGPT Space is genuinely the better choice

This section is not a formality.

Where you need something else

Looking for a ChatGPT Space alternative?

It depends on which half of Space you want to replace, and the honest answer is different for each.

If you want an alternative for team documents, decks, and Sites written with AI, Busabase is not it. Stay with Space or look at another document suite; that is not the job we built for.

If what you need is a place where agent-produced data is structured, reviewed, and usable by any agent — with a proposer, reviewer, and history on each record, open source, and able to run on your own machine — then Busabase is an alternative for that job. More often it is a complement: keep Space for the documents, and put the records your agents produce in Busabase.

What Busabase does differently

Material writes can arrive as a Change Request rather than a silent mutation. The proposal carries a field-level diff, the agent that proposed it, and the source. A person opens it, sees exactly what would change, and approves, comments, or rejects. Only then does it become canonical — and proposer, reviewer, commit, and history stay attached to the record afterwards.

Three things this is deliberately not:

It is not review on every write. Review is proportional to consequence: a low-risk operation stays fast, and an agent whose key has merge permission merges immediately. The credential’s permission level decides, not a blanket rule.

It is not tied to one agent vendor. The same Base is reachable over MCP, an Agent Skill, and OpenAPI, so the agent that proposes a value and the agent that reads it next can come from different companies.

It is not only a hosted service. Busabase is MIT-licensed and local-first: npx busabase server starts a workspace on your own machine, and Busabase Cloud is there when you want it hosted.

Using them together

For many teams the most useful arrangement is both. People and Dots draft, discuss, and present in Space. When a conclusion has to become a record — a supplier decision, an account update, a price, a policy exception — ChatGPT proposes it into Busabase over MCP, the owner reviews the field-level change, and every other agent reads the accepted value from there.

There are two reviews in that chain, and they protect different things. ChatGPT’s app confirmation protects the tool call; Busabase’s Change Request decides whether the fields become canonical. How to connect ChatGPT to Busabase.

Honest limitations

Busabase is younger and smaller than anything OpenAI ships. It has no real-time co-editing to Space’s standard, no presentations, and no built-in agent that goes off and does work for you. If you are judging the everyday experience of writing with AI, Space wins, and it is not close.

Busabase is also not an application backend — not OLTP, not high-frequency machine writes, not vector recall. If that is the need, you want Postgres or a vector store.

And this page was written one day after launch. OpenAI moves fast; several items marked “not described” may change, and we will update the page when they do.

Frequently asked questions

Next step

If you want to see the write path rather than read about it, start a workspace on your own machine with npx busabase server, connect ChatGPT or the agent you already use, and watch one Change Request arrive.

Connect ChatGPT to Busabase · Connect another agent · Browse templates · Read the docs · What makes a store a system of record for AI agents

Sources — OpenAI: ChatGPT Space overview, Pages in Space, ChatGPT Space for Enterprise. Launch coverage (29 September 2026): VentureBeat, TechCrunch. Statements about Dots and auto-review come from launch coverage, not OpenAI’s documentation. OpenAI ships frequently; this page describes what was published as of the verification date above.