Limits and allowances
Every limit OMAI actually enforces, with its real number and what happens when you reach it — AI answers per plan, fair use, team seats, file sizes, text lengths, rate limits, time windows and retention.
- Who
- Owners, admins and anyone who has just hit a wall
- Plan
- All plans — the numbers differ per plan
- Role
- None to read; changing a plan needs Owner or Admin
Everything below is enforced by the product: reach the number and something is refused, paused or stopped. The last section lists figures you will see in the interface that OMAI does not enforce — Meta’s own limits and illustrative estimates — so you can tell the two apart.
What each plan allows
| Plan | AI answers included | Answering stops at | Team members | Knowledge storage | Dedicated numbers you can buy |
|---|---|---|---|---|---|
| Starter | 1,250 | 1,250 | 2 | 125 MB | 1 |
| Growth | 4,000 | 6,000 | 5 | 400 MB | 1 |
| Pro | 10,000 | 15,000 | 15 | 1,250 MB | 3 |
| Custom | Agreed per contract | Agreed per contract | 100 | 10,000 MB | 10 |
Starter stops at exactly its included amount. Growth and Pro keep answering past the included amount, up to half as much again — the fair-use band — and stop there. The billing page shows your own stopping point in the Answers before service pauses tile, and warns you on the way: a banner appears once you pass 80% of the included amount, and again at 100%. When answering stops, incoming messages still arrive and are stored, and each new conversation is handed to your team in the inbox. An organization with no subscription at all is allowed 1 team member, not 2.
Limits that pause the assistant
| Limit | Value | What happens when it trips |
|---|---|---|
| Assistant replies in one conversation, per hour | 20 | That conversation is set to AI paused. It stays paused until someone chooses Resume AI in the inbox. |
| Assistant replies across the organization, per hour | 300 | Further replies are held back for the rest of the hour. No conversation changes state, and the assistant resumes on its own. |
| Assistant reply length | 900 characters | A longer generated reply is not sent. The conversation is handed to your team instead. |
Files and the knowledge base
| File type | Maximum size | Other limits |
|---|---|---|
| 20 MB | Up to 300 pages. A scan with no text layer is rejected — OMAI does not read text out of images. | |
| Word (DOCX) | 20 MB | Files containing macros are rejected. |
| Text (TXT) | 5 MB | — |
| Markdown (MD) | 5 MB | — |
| CSV | 10 MB | Up to 50,000 rows. |
The upload form states the limit for the file type you have selected, and changes it when you change the type — so the figure on screen is the figure that is enforced, and a file over it is refused in the form rather than accepted and then rejected. One file at a time: dragging several onto the drop zone takes the first. A file name may be up to 255 characters, and the Title field up to 200. The number of sources you may keep is not capped; what is capped is their total size, against your plan’s knowledge storage. Documents you delete stop being used immediately and free their space.
Text length limits
| Field | Limit | Enforced |
|---|---|---|
| Reply in the inbox | 4,096 characters | Browser and server |
| Tag on a conversation | 40 characters | Browser and server |
| Internal note | 2,000 characters | Browser only |
| Value for a template variable | 640 characters | Browser only |
| Template name | 120 characters, letters, digits, spaces, dots and hyphens | Browser and server |
| Template body | 2,000 characters | Browser and server |
| After-hours message | 1,000 characters | Browser and server |
| Message in the test simulator | 4,096 characters | Browser only |
| Question in the retrieval test | 500 characters | Browser only |
| FAQ question / answer | 500 / 4,000 characters | Server |
| Tags on one FAQ | Up to 20 tags, each up to 40 characters | Server |
| Service name / description / eligibility | 160 / 2,000 / 500 characters | Server |
| Policy title / body | 200 / 8,000 characters | Server |
| Business name / description / address / city | 160 / 2,000 / 300 / 120 characters | Server |
How much a screen loads
| Screen | Rows loaded |
|---|---|
| Inbox conversation list | 100 most recently active conversations |
| Messages inside an open conversation | The whole thread, however long it is |
| Leads | 200, ordered by score then date |
| Invoices on the billing page | 50 most recent |
| Audit log | 200 entries by default, 500 at most |
| Templates pulled by Sync from provider | 100 per sync |
Time windows
| Window | Length | What happens at the end |
|---|---|---|
| Free trial | 14 days | The trial ends at the moment it was due to end, with no grace period, and the assistant stops answering until a plan is paid for. |
| WhatsApp customer-service window | 24 hours from the customer’s last message | Free-form replies can no longer be sent — by you or by the assistant. The composer switches to approved templates. |
| Guided check session | 60 minutes | The session expires and you start a fresh one. |
| Waiting for a phone-side approval | Up to 10 minutes, rechecked every 8 seconds | The connect card stops waiting and reports that the connection failed. You can start again. |
| Quiet window after you reply from your phone | 6 hours by default, settable between 1 and 24 | The assistant resumes answering that conversation by itself. |
| Settling hold after a phone number is connected | Up to 24 hours | While contacts and chat history are still arriving, the assistant holds back on a number set to answer new contacts only. |
| Checkout link | 24 hours | The payment link stops being valid and you start the checkout again. |
| Waiting for a payment confirmation | 3 minutes | The success page switches to “still waiting” and offers Check again. You are not charged twice. |
| Retries after a failed renewal | 1, 3 and 7 days after the due date, 4 attempts in total | After the last attempt the subscription becomes Unpaid: the assistant stops answering, your team can still reply. |
| Unpaid subscription | 30 days | The subscription is canceled. |
| Numbers rented from OMAI while unpaid | 14 days | The rented number is released. |
| Sign-in session | 30 days | You are asked to sign in again. |
| Recovery window after requesting deletion | 24 hours | Deletion proceeds. Write to us inside the window if you change your mind — there is no cancel button in the dashboard. |
| Backup copies after a deletion | Up to 35 days | The backups holding the deleted data expire. |
Rate limits you can actually hit
These count actions per organization in a rolling minute. Going over does not break anything — the action is refused with a “too many attempts” message and works again a minute later.
| Action | Attempts per minute |
|---|---|
| Starting a checkout from a plan card | 20 |
| Submitting the checkout details form | 10 |
| Starting a payment-method update | 10 |
| Confirming authority on the connect card | 30 |
| Starting the guided check | 20 |
| Running the guided check | 60 |
| Checking whether a phone-side connection completed | 60 |
How long things are kept
Conversations, messages, contacts and knowledge sources are kept for the life of the account: nothing in the product deletes them on a schedule, and there is no setting that shortens it. You remove them by deleting the source, or by deleting the organization. Invoices and the fiscal documents behind them are kept for the statutory period, even after an account is closed.
Numbers that are not OMAI’s to enforce
You will meet these figures in the interface or in Meta’s tools. They are real, but they are not limits OMAI applies, so do not plan against them as if they were.
- Meta’s registration attempts. Meta allows a limited number of registration attempts per number in a rolling 72 hours. OMAI’s protection is to register a number once and never retry automatically; the guidance text tells you to wait for Meta’s window to reset.
- About 20 messages per second on a number that also runs on your phone. That is Meta’s throughput figure, shown in the guided check. OMAI does not measure or shape your sending rate.
- Up to about six months of recent chats imported when you connect a number that also runs on your phone. Meta decides what it sends and how far back it goes; the import is best effort.
- The per-send cost shown on the Templates page is an illustrative default for a template message, not a contracted price, and not what Meta will invoice you.
- Connected WhatsApp numbers per plan. Pro and Custom include more than one connected number; Starter and Growth are sold as single-number plans. The connect flow does not stop you at that line, so treat it as a plan term rather than a wall.
- Rate limits are counted by each web server separately, so the effective number can be a little higher than the table above, and counters reset when a server restarts.
Can I buy more AI answers without changing plan?
No. There is no per-answer purchase and no top-up. Above the included amount, Growth and Pro use the fair-use band at no further charge, and past the stopping point the way to get more answers is a larger plan.
Do messages my team writes count against the allowance?
No. The allowance counts AI answers only. Replies your team types, incoming customer messages, template sends and messages in the test simulator are not counted as AI answers.
What happens to customers when the allowance runs out?
They can still message you and nothing is lost. The assistant stops replying, and each conversation is parked for a person, so it sits in the inbox waiting for your team.
My upload was refused but the file looks small enough.
Check the type. Text and Markdown files stop at 5 MB even though the form says 10 MB. A PDF over 300 pages, a CSV over 50,000 rows, a password-protected PDF and a document containing macros are all refused as well.
