OpenClaw Update Recovery
Use this page when an updater finished with errors, the Gateway will not start, or Doctor claims agent data is damaged. Normal scheduled restarts are a different topic.
Do not delete the data directory first. v2026.10.9 exists because several recovery paths used to treat intact data as lost. Official notes: v2026.10.9.
Copy the evidence
- Write down
openclaw --versionif the binary still runs. - Copy the config/data directory reported by Doctor to a second disk.
- Save updater output and
openclaw logsfrom the last boot.
Match the symptom
| What you see | What 10.9 changed | What you do |
|---|---|---|
| Doctor migration hangs or never starts after an update; huge conversation archives | Archives are processed one record at a time, not expanded into one giant string. Unchanged archives keep original bytes. | Install a build that includes 10.9, then re-run Doctor. Do not expand the archive by hand unless official recovery commands say so. |
| Doctor says agent data is damaged after a Node or SQLite upgrade; search index looks stale | Eligible intact databases can be backed up and repaired in place. Stale indexes are not automatically “destroyed data.” | Follow the printed update and recovery guidance. Activation-time backup support needs an updater that already contains the fix. |
| Git checkout update failed; untracked files or local edits missing after rollback | Failed Git updates now preserve detected edits and untracked files instead of deleting them during rollback. | If rollback is refused, save your changes before running the printed recovery commands. The updater still tries to restore the previous runtime. |
| Windows package update rollback fails on file locks | Rollbacks retry temporary locks. If the package is still missing, the error names retained recovery folders. | Close processes holding those files, then move folders as the error describes. |
| Same-version update or repair left the Gateway stopped even though the core is unchanged | 10.9 restarts when data and ownership checks allow it. The original update failure is still reported. | Fix the reported failure. Recovery stays blocked when checks do not permit a restart. |
| Disk was full; agents refuse new writes | After you free space, agents can accept writes without restarting the Gateway once cleanup and integrity checks succeed. Interrupted writes are not replayed. | Free disk, wait for cleanup, then retry. Do not expect a half-written message to come back. |
Then upgrade the updater
Several of these repairs only run if the installed updater already contains them. If you are stuck on an older 10.x, get to 10.9 with a method that does not wipe the data directory (reinstall the package next to existing config, not a “clean” wipe). Planned upgrades: Upgrade & Migration.