APIs and integrations
Slack API Rate Limits: How the Tiers Work and How a Client Should Back Off
In short
Slack API rate limits are counted per method, per workspace, per app, so one throttled call should pause only that method for that workspace, not your whole integration. A 429 carries Retry-After in seconds. Since 29 May 2025, new installs of unlisted commercial apps get 1 request a minute on conversations.history and conversations.replies.
How are Slack API rate limits scoped, and what are the tiers?
Slack evaluates Web API requests per method and per workspace, inside one-minute windows. The overview adds the app as a third key, so the same method can be throttled in one workspace and free in another. Every Slack plan gets the same tier for a given method.
Each method's reference page lists its tier in a Facts block. The tiers are floors with a plus sign, and Slack says each one is subject to change. Methods with cursor pagination only get their stated tier when you paginate; fetching a full list such as users.list without a cursor gets a stricter allowance.
The table below groups the separate counters a typical bot touches. They are independent, so being under one says nothing about the others.
| Counter | Documented allowance | Keyed on |
|---|---|---|
| Web API Tier 1 / 2 / 3 / 4 | 1+, 20+, 50+ and 100+ requests a minute | Method, workspace, app |
| Special tier (for example chat.postMessage) | Set by the method; about one message a second per channel plus a workspace-wide cap | Channel and workspace |
| Incoming webhooks | 1 per second, short bursts allowed | Webhook |
| users.profile.set | 10 updates a minute for one user's profile; 30 profiles a minute per token | Token |
| Events API deliveries | 30,000 per 60 minutes | Workspace and app |
Checked against: source 1
What changed in 2025 for apps outside the Slack Marketplace?
A changelog entry dated 29 May 2025 moved conversations.history and conversations.replies from Tier 3 to Tier 1 for apps that are commercially distributed but not approved for the Marketplace, which Slack calls unlisted apps. For those apps the two methods allow 1 request a minute, and the default and maximum for the limit parameter dropped to 15 objects.
Who is hit depends on dates and install type. New unlisted apps and new installations of existing unlisted apps got the lower limit immediately. Existing installations were left alone. Internal customer-built apps are excluded and keep 50+ requests a minute with up to 1,000 objects per call; Slack later edited the entry to make that clearer.
The same entry updates the API Terms of Service: effective at once for apps created on or after 29 May 2025, and from 30 June 2025 for older apps. If a new customer install suddenly crawls through channel history, check this before you debug your own code.
Checked against: source 2
What should a client do with a 429 and Retry-After?
Every HTTP-based Slack API, incoming webhooks included, answers an exceeded limit with HTTP 429 and a Retry-After header holding a number of seconds. Slack's own example is a 30-second wait on conversations.info. The wait applies to that method for that workspace, for every token your app holds there.
Other methods for the same workspace stay open, and so does the same method in other workspaces. A single global sleep therefore wastes capacity. I would key the backoff on the pair of method and workspace and keep everything else flowing.
- Read Retry-After on every 429 and park only that method and workspace for that many seconds.
- Paginate list methods with cursors so the documented tier applies at all.
- Pace each call type near one request per second; Slack tolerates short bursts above that but will not publish burst numbers.
- Queue outgoing messages per channel at about one a second, since bursts may never be stored or shown to users.
- On the legacy RTM websocket, stop writing after an error reply; continued sending leads to disconnection and can get the app disabled permanently.
Checked against: source 1
How does the Events API limit and retry deliveries to your server?
Slack caps event deliveries at 30,000 per workspace per app in 60 minutes. Past that, it sends an app_rate_limited event once for each minute you are limited, with a minute_rate_limited epoch value and the team_id involved. Those events are your signal that some workspace activity was not delivered.
Your endpoint must answer with a 2xx within 3 seconds. A failed delivery is retried up to 3 times: almost at once, after 1 minute and after 5 minutes. Each retry carries x-slack-retry-num and an x-slack-retry-reason such as http_timeout or http_error, and retries count toward the failure limit.
If more than 95% of attempts fail within 60 minutes, Slack temporarily disables the app's subscriptions and emails the owner; apps receiving under 1,000 events an hour are not disabled automatically. Send x-slack-no-retry: 1 on a non-200 reply to drop one event. The optional Delayed Events setting adds hourly retries for 24 hours; without it, Slack does not deliver an event more than two hours late.
Checked against: source 3
Does Socket Mode change the delivery picture?
Socket Mode swaps your public request URL for a WebSocket URL that you obtain from apps.connections.open, a Tier 3 method. An app can hold up to 10 connections at once, and any payload may arrive on any of them, so do not route work by connection.
You still acknowledge every event by sending back its envelope_id, because that acknowledgement is how Slack decides whether to retry. Connections are refreshed every few hours, sometimes with a disconnect warning about 10 seconds ahead. A second open connection keeps events flowing during a refresh.
The Socket Mode guide does not state a separate delivery ceiling, and the Events API page does not exempt it. I would plan for the same per-workspace cap and treat Socket Mode as a transport change, not a capacity change.
What do Slack's pages leave unsaid?
Exact burst sizes are deliberately unpublished, and tier floors can move. The workspace-wide cap on chat.postMessage has no number on the pages read for this post. Slack also reserves the right to rate limit other functionality without a descriptive error, so log the method name and workspace with every 429 you see.
Checked against: source 1
What else do people ask?
Does a Slack 429 block my whole app?
No. The documented scope is one method for one workspace. Other methods in that workspace, and the same method in other workspaces, keep working.
Are Marketplace apps affected by the May 2025 change?
Slack's changelog says Marketplace apps should see no rate limit change, and internal customer-built apps are excluded too.
Sources
- Slack Developer Docs, Rate limits, fetched 2026-11-20
- Slack Developer Docs changelog, Rate limit changes for non-Marketplace apps (29 May 2025), fetched 2026-11-20
- Slack Developer Docs, The Events API, fetched 2026-11-20
- Slack Developer Docs, Using Socket Mode, fetched 2026-11-20
- Slack Developer Docs, apps.connections.open method, fetched 2026-11-20
How was this page made?
I drafted this page with AI assistance from the official documents listed under Sources. This page compares Slack's own developer documents, fetched on 20 Nov 2026. I did not run a Slack app or send requests for it, so nothing here is a measurement. It does not cover the SCIM, Audit Logs or Discovery APIs, workflow trigger limits, or per-method numbers beyond the ones quoted. Found a mistake? Tell me.