Supabase Realtime has four separate caps, not one
A Realtime problem on Supabase is rarely "too many connections" in the abstract — it's one of four specific, separately documented ceilings: how many sockets can be open at once, how many messages per second can flow, how many channel joins per second are allowed, and how big one broadcast payload can be.
Last checked November 18, 2026.
The four caps, by plan
| Cap | Free | Pro |
|---|---|---|
| Concurrent connections | 200 | 500 |
| Messages per second | 100 | 500 |
| Channel joins per second | 100 | 500 |
| Broadcast payload size | 256 KB | 3,000 KB |
What hitting each one looks like
- Concurrent connections: every open browser tab subscribed to a channel holds one socket. A dashboard app left open across a team, or a mobile app that reconnects aggressively on flaky networks without closing the old socket first, burns through 200 faster than raw user count suggests.
- Messages per second: a busy presence channel (many users moving a cursor, typing indicators) generates far more messages per second than it looks like from the UI, since every state change broadcasts to every subscriber.
- Channel joins per second: a page that subscribes to a fresh channel on every route change, instead of reusing one long-lived subscription, can hit this during a traffic spike even with modest total user count.
- Broadcast payload size: sending a large JSON blob (e.g. a full object instead of just the changed field) over broadcast hits the per-message size cap before the connection or rate limits do.
The fix
- Close subscriptions you no longer need instead of letting them pile up across route changes or reconnect attempts — reuse one channel per logical feed rather than opening a new one per component mount.
- Send only the changed fields over broadcast, not the whole row or object, to stay well under the payload cap and reduce bandwidth.
- For high-frequency state like cursors or typing indicators, throttle/debounce on the client before broadcasting rather than sending on every event.
- If real usage genuinely needs more than the Free caps, Pro raises every one of the four ceilings (200→500 connections, 100→500 messages/s, 100→500 joins/s, 256KB→3,000KB payload).
Scribase realtime, same underlying idea
Scribase gives every project realtime on Postgres: subscribe to table changes, broadcast messages, or presence, and update the UI when an event arrives — the same client-side subscription model Supabase uses. This page does not claim a specific published connection ceiling for Scribase the way the table above states for Supabase; the fixes above (closing unused subscriptions, sending only changed fields, throttling high-frequency events) apply the same way regardless of host.
Sources: Scribase pricing
Questions people ask
Which of the four limits do real apps hit first?
Concurrent connections, most often — it accumulates from every open client, including ones left idle or reconnecting repeatedly, rather than scaling cleanly with active user count.
Does Pro remove these limits entirely?
No — Pro raises all four numbers (connections, messages/s, joins/s, payload size) but each one is still a documented, fixed ceiling, not unlimited.
Is a slow typing-indicator feature likely to hit the messages-per-second cap?
Yes, if every keystroke broadcasts its own message to every subscriber — debouncing on the client before sending is the usual fix rather than raising the plan tier.
Does closing a browser tab free its connection immediately?
Usually, but a flaky network can leave a socket half-open briefly before the server detects the disconnect — aggressive client-side reconnect logic that does not clean up the old socket first is a common way this compounds.