Limits

The Change it column says how to go past a limit:

• You — a setting you change yourself (the link shows where).
• Upgrade — a higher plan has a higher limit (plans and prices).
• Ask us — Totalum can raise it for your account or project: write to support with what you need and why.
• No — fixed for everyone.

Plans are named Free, Starter, Builder, Professional and Enterprise. All amounts are per calendar month (UTC) unless the row says otherwise.

#Account

#Plan limits

• A deleted project frees its slot at once. Creating too fast answers 429 RATE_LIMIT_EXCEEDED; wait a moment and retry.
• New projects and prompts share one per-minute allowance: a prompt over it, or over the prompts running at once, answers 429 RATE_LIMITED with Retry-After; nothing is started, so send it again after that many seconds. A prompt stops counting as running as soon as it finishes, fails or is stopped. If your account has a higher new-project rate from us, its prompt limits rise with it (running at once = twice the per-minute figure).
• A key over its requests per second answers 429 RATE_LIMITED with Retry-After (counted each second, Enterprise over 10 seconds: spread a burst over a few seconds).
• A pending invitation holds a seat until it is accepted, revoked or expires.
• Code editing, source download, GitHub, custom domains and mobile apps need a paid plan; reading your code and connecting Figma work on every plan.
• The API and MCP work on every plan. Any paid action needs a balance of at least 1 credit (reads and deleting a project always work).
• The API key the platform gives an app-builder project (handbook): 600 requests a minute and 300 prompts an hour (Ask us); its project creates count against the account's per-minute rate above.

#Account spending caps

To protect your balance from a runaway app or script, the whole account pauses when everything it spends — prompts, publishes and every project's running app — passes a cap:

While your balance is 0 or below, the caps are at most 100 credits an hour and 500 a day. The pause lifts by itself once the spend is back under the cap and the balance is above 0. Your caps are in Get Account Usage (account.accountCaps).

#Balance and suspension

Hosting, the database and other infrastructure keep running when the balance reaches 0, so a balance can go below 0. Then:

Nothing is deleted. Add credits (or turn on auto-recharge) and the sites come back by themselves within a minute.

#Project

#Monthly budgets

Every project has two monthly caps, in credits: development (prompts, publishes and project actions) and infrastructure (what the running app spends). New projects get:

Set them at creation (creditLimits in Create Project) or later in the project's Usage & limits or with Update Credit Limits. You can also switch any built-in feature off or give it its own budget (Feature Limits).

#Built-in features

What each project's app may use. Each cell is per month · per minute.

• Change it: Upgrade. Monthly amounts are per project. On Free, Starter and Builder the per-minute amounts are shared by all the account's projects (except file uploads); on Professional and Enterprise each project has its own.
• Over a limit, the app's call answers 429 with Retry-After (per minute) or errorDetails.resetsAt (per month). Every call also costs credits, so the budgets above usually stop spending first.
• AI text answer length: up to 4,096 tokens per call on Free, 32,768 on Starter, Builder and Professional; no ceiling on Enterprise (32,768 when the call does not ask for one). A call that names no length gets that amount or the model's own maximum, whichever is lower. Upgrade.
• Cloud browser: 1 session open at a time on Free, 3 on paid plans (Ask us); a session lasts up to 15 minutes on Free, 60 on Starter, 180 on Builder and 360 on Professional and Enterprise (Upgrade).
• Web scraping: 1 call at a time per project; a call waits in a queue for its turn (Ask us).
• All built-in calls together: 100 a second per project (Ask us).
• Per visitor of the published site: 20 paid built-in calls a minute (AI, web, browser, screenshots), so nobody can spend your credits by calling a public AI page in a loop — ask the agent to change it (You); and 5,000 requests a minute, adjustable from 10 to 60,000 on paid plans with Update Rate Limit (You).

#Emergency stop

A project whose app suddenly uses far more than normal is paused: its site shows a paused page, its built-in calls answer 429 PROJECT_PAUSED_RUNAWAY, and you get an e-mail and the project.kill_switch.fired webhook. It trips on any of:

It also trips on unusually costly use of built-in features within one hour. Prompts never count, nor does the hour's single largest call (one long video paid up front).

• Halfway: once the app has spent half of its hourly credits above, its image, video and speech AI calls answer 429 AI_MEDIA_PAUSED_THIS_HOUR until the hour ends; text AI and everything else keep working.
• Lifting it: the pause lifts by itself after 24 hours. Fix the cause, then lift it at once with Bring it back online in the project's workspace or with Release Runaway Pause (free, also clears the halfway pause).
• Change it: Upgrade. The thresholds are not raised per project; releasing the pause is in your hands.

#Database

Each project has a development and a live database (SQL, SQLite on Cloudflare D1).

How to write code that fits them: Design for the Limits.

#Files and storage

Files uploaded through the API: 20 MB each and 300 a minute. Backups count as storage.

#Real-time

Change it: Upgrade. One room handles about 1,000 messages a second in total.

#Scheduled jobs

A job pauses after 3 failed runs in a row, unless it sets its own number.

#Domains

A domain's first year can cost less than its yearly renewal: both prices are shown before you buy.

#Agent, prompts and code

All API request limits: Errors & Limits.

#Idle projects

Any project wakes from a short idle period by itself on its next prompt.