Roblox limits how many requests one internet address may send to its public endpoints. The collector and the API share the studio's address, so those limits are one budget for everything the platform reads from Roblox. Spending it carelessly means refused requests, and readings that are never taken cannot be taken later.
Why requests to Roblox are limited
The endpoints the collector reads are the public ones Roblox's own website calls, limited per address. Going over makes Roblox answer 429 (too many requests) and block that endpoint for a while. During a block no game is read, and since Roblox serves no history, those readings are lost for good. So the platform stays under every limit by design, and treats new requests as a cost to justify.
The hidden cap on game details: how many games can be tracked
Game details (players, visits, favorites, genre) are the reads that matter most, and their real limit is not the published one. Roblox's header says 300 requests a minute, but measured on 2026-09-29 it refuses after about 550 games a minute, however the requests are spaced: the cap counts games, not requests. The collector sends 10 requests of 50 games a minute, 500 games a minute, leaving one batch of margin.
That one number sets how many games the platform can track: 500 a minute is 720,000 snapshots a day, divided between the tiers.
How the budget is shared
Each process has its own limiter with a bucket per endpoint:
The collector takes half (0.5) of each published limit.
The API takes a fifth (0.2,
ROBLOX_RATE_LIMIT_FRACTION) per copy, and caches what it reads.Game details, whose limit was measured, are not divided: each process may send 10 batches a minute, so a lookup by the API while the collector sends at full rate spends games from the same hidden cap.
Together the collector and the API's two copies take nine tenths (0.5 + 0.2 + 0.2) of each published limit, leaving a tenth for error.
When more games are due than the budget allows, the collector reads hot games first, then active, warm, cold and dead, and the Collector page shows them as overdue.
The endpoints and their limits
| Endpoint | For | Batch | Published limit |
|---|---|---|---|
| Game details | Players, visits, favorites, genre | 50 | 300 a minute, about 550 games a minute in practice |
| Votes | Upvotes and downvotes | 100 | 200 a minute |
| Icons | Game icons | 100 | 1,200 a minute |
| Explore sorts | Discovery sorts and ranks | 1 page | 120 a minute |
| Recommendations | 6 games recommended beside a game | 1 | 200 a minute |
| A group's games | A creator's other games | 1 | 3 a second |
| A user's games | A creator's other games | 1 | 200 a minute |
| Place to game | Matching Rolimon's places | 1 | 60 a minute |
The full findings are in the vault note Roblox Endpoints.
What opening an untracked game costs
Opening a game's page that nobody tracks yet (GET /v1/games/:id) asks Roblox for that one game: one game-details request carrying one game. It is one of the API's 24 lookups a minute (120 times 0.2, per copy) and one game of the roughly 550 a minute the address allows. If the game exists, it is then tracked and read on schedule from then on, which is the larger cost.
Each such lookup counts against your external limit, 20 a minute per person and 40 per address, before Roblox is asked, so one person cannot spend the API's whole share. Opening a game that is already tracked costs nothing. The assistant asking about a game never starts tracking it.
What group and key actions cost
These are an owner's actions, a handful a month, and never on a schedule:
Adding a group, replacing its key or checking it: one key introspection and one keyed group read on Open Cloud (limited per key, not per address), plus the group's public profile and icon (two requests on the API's share).
Registering a private game: one keyed game read, plus the public lookup.
Listing a group's games to register: up to 5 pages of the group's games, one details batch of 50 and up to 3 icon batches, cached 10 minutes.
What owned analytics cost
Owned analytics are read with each group's own key, and Open Cloud counts those per key, so they never touch the address's budget. Per studio game and daily read: 35 analytics queries (28 daily totals, revenue by source, 6 breakdowns), up to 6 status checks per query Roblox is still preparing, and one page each of developer products, game passes and bans for a typical game. Each kind has a bucket of 30 a minute in the worker. About 175 queries a day for 5 games, 700 for 20.
What costs nothing
Every page of the panel, the Radar, growth figures and notifications: they read stored data.
Your dashboard, since 2026-10-22: it lists your projects' games from the database.
Opening a tracked game, its history, sorts and recommendations.
For developers: adding a request
A new or faster request to Roblox spends the same budget. The collector's schedules and bucket sizes are not changed without these steps.
Read the endpoint findings
Read the Roblox Endpoints note in the vault: the endpoint's real limit, batch size and how it fails.
Use the shared client
Add the request to the shared client in
packages/roblox, with its own limiter bucket, timeout and retry rules.Add its fake
Give the test fakes in
packages/testingthe same endpoint, so no test calls Roblox.Say what it costs
In the pull request, write what it costs per minute and per day, and what it leaves of the budget.
Related: Collector and freshness, Tiers and genres.
Updated