Storage and Limits

Each ObjectOS Cloud environment has a few limits of its own. This page lists them with the plan each number belongs to. The AI allowance has its own page: AI Credits and Limits.

Storage

Every environment has a storage allowance: 1 GB on Free and 20 GB on Team, counted per environment. A development environment gets the same allowance as production. An Enterprise contract carries no storage number. (Product owner ruling, 2026-09-08: the numbers are final.)

What counts. The environment's database plus its attachments. Usage is re-measured about every hour, so a large upload or import shows up in the reading within the hour; the lag is deliberate.

What a full environment stops. From 100% of the allowance, two things are refused:

• attachment uploads (file and image fields);
• bulk import of records.

The refusal is 403 with the code STORAGE_LIMIT_REACHED and the way to upgrade.

What it never stops. Creating, editing or deleting a single record; reading anything; exporting; and deleting attachments to free space. Someone using an app built on the environment is never left unable to save a form because the environment is near its allowance.

The warning. At 80% the console shows a banner. It is a warning, not a limit: uploads and imports still go through until 100%.

Downgrading deletes nothing. An environment that ends up over the Free allowance after a downgrade keeps every byte, stays readable and exportable, and only stops taking uploads and imports until it is upgraded again or space is freed.

MCP limits

Every environment's MCP endpoint accepts a set number of requests, shared by every agent connected to that environment. Ruled on cloud#3203.

• Free: 20 requests a minute and 1,000 a UTC day.
• Team and above: 120 requests a minute, with no daily cap.

A request is one MCP message: a tool call, a listing, each step of the connection handshake. A batch counts each message in it, and a batch the day's remaining count cannot hold is refused whole. The per-minute allowance refills continuously, so a short burst up to the per-minute figure is fine.

Past a limit the endpoint answers 429 Too Many Requests with a Retry-After header, and the body's error.code is RATE_LIMIT_EXCEEDED. The message says which limit was hit and when it resets; on Free it also points to the upgrade to Team. A refused request did not run and uses up neither limit, and the next one after the wait is served normally. An agent driven by a person stays well under the paid limit; one that hits it is usually a loop calling the same tool over and over.

Drafts and authoring writes

Changes to an app's design are staged as drafts until they are published. An environment holds at most 500 unpublished drafts, on every plan and whichever way they were made. Past that, a change that would stage another draft is refused until drafts are published or discarded; no clock resets this one.

Authoring writes that arrive through MCP — the build tools that create or change metadata — are also counted per UTC day: 200 on Free and 2,000 on Team and above. Reads and checks are not counted. Both refusals use the same RATE_LIMIT_EXCEEDED answer as the request limit, with the count, the limit and when it resets. Ruled on cloud#3303.

Next

• Plans and AI Seats — what each plan includes.
• Connect AI Tools (MCP) — turn on an environment's MCP endpoint.