# Capacity: per-workspace rate limits and smoother schedules

> An audit of what would break at ten times our usage lifted three limits: per-workspace rate limits, non-blocking token refreshes, and spread-out schedules.

Sep 24, 2026. Topics: Platform. https://www.generalinput.com/changelog/2026-09-24-capacity-improvements

We audited what would break if usage grew tenfold, sized each fix from measured numbers, and lifted three limits that were already slowing busy workspaces down. The result is steadier performance at the moments workspaces are busiest, like the top of the hour when many schedules fire at once.

### How it works

- **Per-workspace rate limits.** Calls from workflows to our built-in services shared one rate limit across every workspace. Each workspace now gets its own 1,000 requests a minute, so one busy workspace can't slow down another.
- **Token refreshes no longer block the database.** Refreshing a connected account's token used to hold a database connection during the provider's response. A token that's still valid now skips the lock entirely, waiting requests release their connection while they wait, and provider calls time out after 30 seconds.
- **Schedules spread out.** Each scheduled trigger starts at a small, stable offset, so a burst of runs on the hour no longer spills past the pool of warm sandboxes.
- **Faster long conversations.** Looking up the latest message in a long conversation dropped from 67 ms to 0.4 ms.

### Also shipped

- Saving an unchanged schedule no longer rewrites it, which used to drop its next pending run.
- 31 read-only triggers now retry temporary server errors.