When Grok Bot routine create returns Not saved

Creating or updating a routine can fail with:

Not saved — the routine could not be saved on the server this turn (create failed); try again in a moment.

The same string appears for update and delete failures. Forum 171895: staff (deanrie) confirmed a fleet-wide routine save outage on Sep 17 2026 (~6:30–10:40 PT) that was fixed the same day. On Sep 25, users again reported create failures on Grok Bot 0.58 Team while memory writes still succeeded (Matt_Devaney and a colleague). Staff had not posted a Sep 25 root cause at the time of this how-to. Treat a fresh Not saved as a server-side save miss first. Blame a broken Bot profile only after saves keep failing across several Bots and status.cursor.com is quiet.

Confirm it is the save error

1. Ask the Bot to create a small cron routine (simple name, weekday morning schedule, prompt that only posts in that conversation).
2. Note the exact tool error. The create-failed / update-failed / Not saved wording is this path.
3. Check whether a confirm card appeared. Sep 25 reports failed before save, with no confirm card.
4. Ask the Bot to write a short memory fact with update_state target memory. If memory succeeds while routine create fails, you are on the routine-save miss, not the older brain-docs memory bug.

What to try

1. Wait a few minutes and ask the Bot to create the routine again (same payload). Transient save misses have cleared without Reset.
2. Open status.cursor.com and note any open Grok Bot / routines incidents during your window.
3. Keep the owning Bot. Do not Update or Reset Agent Computer as the first fix for Not saved alone — staff told users Reset does not fix the related memory path, and Sep 25 reporters saw no change after a full computer update.
4. If create still fails for hours across more than one Bot, post on forum 171895 (or a new bug) with: Grok Bot version, plan, time zone, exact error string, whether memory writes still work, and whether a confirm card appeared.

Workarounds while create is down

1. Keep using routines that already exist and still fire. Pause or edit them only when those actions succeed.
2. Run the recurring job by hand or with a one-off message until create works again.
3. Do not rely on disk-only edits under local automations folders as a live replacement. Users reported staged folders that never became server routines until create succeeded.

For how routines should work when saves succeed, see Create a Grok Bot routine and Test a Grok Bot routine.

Separate open issue: memory brain-docs

An older error, brain docs snapshot was cut short before completing, blocks update_state memory writes on some fleets while settings updates still work. Staff confirmed that path separately and still recommended writing important facts to memory/profile.md or memory/log/YYYY-MM.md until fixed. That is not the same string as Not saved on routine create. If you only see the brain-docs error, use the file append workaround and leave routines alone.

Pitfalls

• Deleting or recreating the Bot because one create failed — try again and check status first.
• Assuming Reset / Update Computer will clear Not saved — reporters and staff guidance say otherwise for this family.
• Treating a Sep 17-style outage as a permanent account configuration problem.
• Editing local automation JSON and assuming the harness will pick it up without a successful server save.