A practical OpenClaw backup strategy for memory, config, and state.
Use ReClaw restore points alongside careful workspace hygiene so OpenClaw memory, auth, config, sessions, and skills can survive broken updates and host moves.
What To Protect
Treat an OpenClaw agent as stateful software. The model can be reinstalled, but the working setup is spread across memory files, workspace instructions, skills, auth profiles, provider state, sessions, scheduler configuration, and project context.
| Area | Why It Matters | ReClaw Role |
|---|---|---|
| Workspace | AGENTS, memory, identity, user notes, local skills. | Create encrypted restore points before risky edits. |
| State directory | Config, credentials, sessions, provider state. | Capture the full OpenClaw archive locally before upload. |
| Restore metadata | Labels, archive names, hashes, versions, expiry. | Keep a restore-ready ledger for humans and agents. |
Baseline Strategy
Use private git for readable workspace files when that is appropriate, and use ReClaw for encrypted off-site restore points that also cover operational state. Do not commit credentials, sessions, tokens, or encrypted secret payloads to a normal repo.
reclaw backup bootstrap
reclaw backup create --label "Baseline restore point"
reclaw backup list --jsonCadence
Pick cadence by rollback window. A personal agent that changes once a day can start on Free. A production-like OpenClaw workspace that changes during the day should use a tighter window.
- Free: one encrypted backup per day for low-change workspaces.
- Plus: one backup every four hours for regular active work.
- Pro: one backup every thirty minutes for high-risk changes.
Before Risky Changes
Create a labeled restore point before OpenClaw updates, config rewrites, auth migration, plugin or skill changes, and memory system experiments.
reclaw backup create --label "Before OpenClaw update"
reclaw backup create --only-config --label "Config before auth change"
reclaw backup show <backup-id> --decrypt-metadataRestore Drills
A backup strategy is not complete until restore is tested. Run a non-destructive restore into a safe directory and verify the archive before you need it during an outage.
Use the restore drill after the first backup and before major migration work.
Agent Instructions
Give the agent a short operational rule instead of relying on memory:
Before changing OpenClaw config, auth, skills, memory files, or upgrade state:
1. Run reclaw backup create with a useful label.
2. Confirm reclaw backup list shows a READY backup.
3. Continue only after the restore point exists.
4. Never print ReClaw tokens or backup passphrases.