# Rate limits

Every request rate limit Sailfork enforces, what a 429 looks like, and how to back off.

## What is not rate limited

Git has no request rate limit, and neither does the REST API outside the routes listed below. Clone, fetch and push as often as you like. What bounds Git is size:

| Limit | Value |
| --- | --- |
| Objects in one push | 250,000,000 |
| Inflated bytes in one push, all objects together | 1 TiB |
| One push over HTTPS or SSH | 3 GiB of pack |
| One file in Git | 256 MiB |
| One file in Git LFS | 100 GiB |
| Git LFS in one repository | 1 TiB |

A history larger than 3 GiB goes up in several pushes. Every size limit is on [Limits](https://sailfork.com/docs/limits).

## How they work

Each limit counts requests in a fixed one-minute window that starts with the first request. Past the limit the route answers `429` until the window ends. Limits are not sent back in headers, so back off when you see a `429` instead of counting calls.

A window counts one of these:

| Counted per | Meaning |
| --- | --- |
| Address | The caller's IP address. Each IPv6 address counts on its own. |
| Account | The person signing in, whatever address they sign in from. |
| Person | The signed-in person, across all their tokens. |
| Token | One access token or MCP sign-in. |
| Everyone | One window shared by every caller of the route. |

## What a 429 looks like

Most routes answer:

```http
HTTP/1.1 429 Too Many Requests
Retry-After: 23
Content-Type: application/json

{ "error": "too many requests, try again shortly" }
```

`Retry-After` is the number of seconds until the window ends. A few routes answer `429` with their own message and no `Retry-After`: code search, forking, pull request presence, abuse reports, and the per-challenge cap on the second sign-in step. Wait a minute before retrying those.

Git LFS answers in the LFS error format, `{ "message": "too many LFS uploads, try again shortly" }`, with `Retry-After`.

## Back off

1. On `429`, wait `Retry-After` seconds, or 60 when it is missing.
2. Retry once. If it fails again, double the wait each time, up to a few minutes.
3. Do not retry in parallel: every retry counts against the same window.

The CLI treats `429` as exit code 9, unavailable, retry later. See [the CLI reference](https://sailfork.com/docs/cli).

## Signing in

| Route | Limit per minute |
| --- | --- |
| Sign-up, `POST /api/auth/sign-up`, and sign-up with GitHub | 5 per address |
| Sign-in, `POST /api/auth/sign-in` | 20 per address, and 10 per account |
| Second step at sign-in, `POST /api/auth/sign-in/second-factor` | 10 per account |
| Request a password reset, `POST /api/auth/password/forgot` | 5 per address |
| Set a new password from a reset link, `POST /api/auth/password/reset` | 10 per address |
| CLI sign-in, device code, `POST /api/cli/device` | 20 per address |
| CLI sign-in, token exchange, `POST /api/cli/token` | 120 per address |

Sign-up with GitHub that hits the limit returns to the sign-up page with an error instead of a `429`. The second step also stops a single code challenge after a few wrong codes; start sign-in again to get a new one.

## OAuth and MCP

| Route | Limit per minute |
| --- | --- |
| MCP server (`/mcp`, `/mcp/readonly`, `/mcp/x/...`), signed in | 600 per token |
| MCP server, signed out or with a token that does not check out | 120 per address |
| OAuth client registration, `POST /api/oauth/register` | 20 per address |
| OAuth consent, `POST /api/oauth/authorize` | 30 per address |
| OAuth token and revoke, `POST /api/oauth/token` and `POST /api/oauth/revoke` | 60 per address, shared |

## Repositories and pull requests

| Route | Limit per minute |
| --- | --- |
| Code search, `GET /api/search` | 30 per address |
| Fork a repository, `POST /api/repos/{owner}/{repo}/forks` | 10 per person |
| Pull request presence, `PUT /api/repos/{owner}/{repo}/pulls/{number}/presence` | 240 per token |
| Object reads, `POST /api/repos/{owner}/{repo}/objects` | 300 per person, or per address when signed out |
| Git LFS upload batches, `POST /{owner}/{repo}/info/lfs/objects/batch` | 60 per person |
| Git LFS verify, `POST /{owner}/{repo}/info/lfs/verify` | 60 per person, counted apart from uploads |
## The site

| Route | Limit per minute |
| --- | --- |
| Explore, the list of public repositories, `GET /api/explore/repos` | 120 per address |
| Resolving names and emails to people, `POST /api/people` | 60 per address |
| Sync connections for the web app, `/api/sync` | 60 new connections per address |
| Service status, `GET /api/status` | 300 for everyone |
| Abuse reports, `POST /api/abuse` | 5 per address |

## Sandbox previews

| Route | Limit per minute |
| --- | --- |
| Any request to a preview link | 600 per address |
| A preview link that does not exist | 20 per address |
| Reporting a preview link | 3 per address |

See [Sandboxes](https://sailfork.com/docs/sandboxes).

## Other 429s

These answer `429` for a reason other than a request rate:

- Resending a verification email within a minute of the last one, with `Retry-After`.
- Starting a sandbox when the org's Drydock hosts are busy or its sandbox quota is in use. See [Sandboxes](https://sailfork.com/docs/sandboxes).

When GitHub rate limits Sailfork during a move or a mirror, Sailfork waits out limits of a minute or less itself and answers `503` with the wait for longer ones.

---

Every docs page: https://sailfork.com/llms.txt
