# Rate limits

> How many requests each satellite's APIs take, what each limit is counted on, what a 429 answer looks like, and the GraphQL and body size limits they share.

Source: https://docs.bifrostorbit.com/orbit/developers/rate-limits/

Each satellite limits its own APIs. The limits are the same on every plan. When you go over one, the API answers `429` with the code `rate_limited`: wait, then try again.

## How the limits count

Most satellites count with a bucket that refills: you can send a burst of requests at once, and the bucket then fills up again at a steady rate. "120, then 120 a minute" means up to 120 requests at once, then two a second. CMS and Insights Orbit count per calendar minute instead.

Limits are counted on our servers as requests arrive. Treat the numbers as what you can rely on, plan to stay below them, and always handle a `429`.

## CMS Orbit

| API | Limit | Counted on |
| --- | --- | --- |
| Delivery API and GraphQL | 60,000 a minute | Each key |
| | 15,000 a minute | Each IP address, for requests from browsers. Requests from a server with a key, or with a `cms_sk_` key, aren't counted on the address |
| Experiment results from browsers | Share the 15,000 a minute | Each IP address |
| MCP server | 60 a minute, or what the agent key is set to, from 1 to 600. Connections signed in with OAuth get 60 | Each agent key or connection |
| Management API | No limit | |

The delivery API sends `X-RateLimit-Limit`, `X-RateLimit-Remaining` and `X-RateLimit-Reset` with every answer, and `Retry-After` with a `429`. See [CMS Orbit's delivery API](https://docs.bifrostorbit.com/cms/developers/delivery-api/).

## Careers Orbit

| API | Limit | Counted on |
| --- | --- | --- |
| Delivery API and GraphQL | 120, then 120 a minute | Each key and IP address |
| Applications, talent pool sign-ups and interview answers | 6, then 2 a minute | Each organisation and IP address |
| Job view counts | 30, then 30 a minute | Each IP address |
| Candidate portal: sign-in codes | 5, then 1 a minute | Each IP address |
| Candidate portal: checking a code | 10, then 5 a minute | Each IP address |
| Candidate portal: everything else | 60, then 30 a minute | Each IP address |
| Management API | 120, then 120 a minute | Each token and IP address |
| MCP server | No limit | |

## Commerce Orbit

| API | Limit | Counted on |
| --- | --- | --- |
| Storefront reads (shop, products, collections) and GraphQL | 120, then 120 a minute, shared between REST and GraphQL | Each shop and IP address. GraphQL with a key and no `organisation` counts on the key |
| Starting a cart, asking for a return | 10, then 5 a minute | Each IP address |
| Checkout, service points, orders, returns, customer profiles and addresses, downloads, gift cards on a cart | 20, then 20 a minute | Each IP address |
| Personal prices | 30, then 30 a minute | Each IP address |
| Sign-in codes, cart codes, account deletion and link requests | 5, then 1 a minute | Each IP address. Sign-in codes are also limited per email address |
| Checking a code, gift card balances | 10, then 2 a minute | Each IP address |
| MCP server | No limit | |

## Travel Orbit

| API | Limit | Counted on |
| --- | --- | --- |
| Trips, departures, guides and destinations, and GraphQL, from a browser or without a key | 120, then 120 a minute | Each booking name and IP address |
| The same from your own server, with an API key | 6,000, then 6,000 a minute | Each API key, and each IP address |
| Prices (`quote`, REST or GraphQL) | 30, then 30 a minute | Each organisation and IP address |
| Making a booking | 10, then 5 a minute | Each IP address |
| Checking a booking | 20, then 20 a minute | Each IP address |
| Traveller portal: sign-in codes | 5, then 1 a minute | Each IP address |
| Traveller portal: checking a code | 10, then 2 a minute | Each IP address |
| MCP server | No limit | |

## Booking Orbit

| API | Limit | Counted on |
| --- | --- | --- |
| Location, services, staff and free times | 120, then 120 a minute | Each location and IP address |
| Holding a time | 20, then 10 a minute | Each location and IP address |
| Booking an appointment or joining a waiting list | 10, then 5 a minute, shared | Each location and IP address |
| Managing a booking: viewing, cancelling, moving, offers | 20, then 10 a minute | Each location and IP address |
| Paying | 20, then 20 a minute | Each IP address |
| Booking assistant | 8, then 6 a minute | Each location and IP address |
| MCP server | No limit | |

## Presence Orbit

| API | Limit | Counted on |
| --- | --- | --- |
| Review widget | 120, then 120 a minute, for all businesses together. A `429` carries `Retry-After: 60` | Each IP address |
| Feedback and survey forms | 5, then 2 a minute | Each IP address |
| Review requests API and MCP server | No limit. A call takes up to 100 requests | |

## Insights Orbit

| API | Limit | Counted on |
| --- | --- | --- |
| Events from browsers (`/insights/v1/collect`) | 120 a minute | Each visitor's IP address |
| Server API (`/insights/v1/server/events`) | 600 a minute | Each key |
| Raw event API (`/insights/v1/events`) | 60 a minute | Each key |
| GraphQL | 60 a minute | Each key |
| MCP server | No limit | |

A `429` from the server API, the raw event API and GraphQL carries `Retry-After: 60`.

## OAuth and the Orbit app

| API | Limit | Counted on |
| --- | --- | --- |
| Registering an OAuth app (`/oauth/register`) in Careers, Commerce, Travel, Booking and Presence Orbit | 20, then 10 a minute. A `429` answers `slow_down` | Each IP address |
| Registering an OAuth app (`/oauth/register`) in CMS Orbit and Insights Orbit | 10 a minute. A `429` answers `slow_down`; CMS Orbit adds `Retry-After` | Each IP address |
| The Orbit app's token and revoke endpoints | 30 a minute. A `429` answers `slow_down` | Each IP address |
| The Orbit app's API | 240 a minute | Each signed-in device |

## What a 429 looks like

REST APIs answer:

```json
{ "error": { "code": "rate_limited", "message": "Too many requests; try again shortly" } }
```

GraphQL APIs answer with `errors`, and the code in `extensions.code`:

```json
{ "errors": [{ "message": "Too many requests. Wait a minute and try again.", "extensions": { "code": "rate_limited" } }] }
```

The CMS Orbit MCP server answers in JSON-RPC, with `error.code` `-32001`.

Wait for `Retry-After` seconds when the answer has it, and a minute when it doesn't. Then try again, waiting longer each time if it happens again. The CMS Orbit SDKs don't retry a `429` themselves: they throw a rate-limit error that carries how many seconds to wait.

## GraphQL and body limits

Every GraphQL API (CMS, Careers, Commerce, Travel and Insights Orbit) takes the same queries at most:

| Limit | Answer when over it |
| --- | --- |
| 20,000 characters | `413`, `query_too_large` |
| 10 levels deep | `400`, `query_too_deep` |
| 300 fields | `400`, `query_too_complex` |

Bodies:

| API | Largest body |
| --- | --- |
| CMS Orbit MCP server | 5 MB |
| Other MCP servers | 1 MB |
| CMS Orbit experiment results | 16 KB |
| Insights Orbit events from browsers | 64 KB, up to 50 events |
| Insights Orbit server API | 256 KB, up to 100 events |

Pages of results are at most 100 items in the CMS and Careers delivery APIs, 250 products in the Commerce storefront API and 1,000 events in the Insights raw event API.

## Stay under the limits

- Cache answers. A website that caches CMS Orbit's pages for a minute makes a handful of requests, however many visitors it has.
- Call from your server where you can, with a key, so your visitors' requests don't all count against one address.
- Use webhooks to hear about changes instead of asking again and again. See [Verify webhooks from Bifröst Orbit](https://docs.bifrostorbit.com/orbit/developers/verify-webhooks/).
- Ask for what you need in one GraphQL query instead of many REST calls.
