#Memory lost

1 messages · Page 1 of 1 (latest)

neat sedge
#

Hey everyone 👋

I’m facing an issue with OpenClaw memory after the recent update and wanted to check if anyone else has experienced this.

I’m currently running OpenClaw inside a VM, and I usually reset the system when I want to switch or change the LLM configuration. Earlier, this workflow worked perfectly fine. Even after resetting the system and setting everything up again, OpenClaw was still able to recall all the automations I had previously created.

However, after the latest update, something seems different.

Now when I reset the system and start the setup again:
• The memory.md file is missing
• OpenClaw is unable to fetch previous automation data
• It behaves like the entire memory state has been lost

Previously, even after resets, the system was able to reconstruct or access the earlier memory, so I never lost my automation setup. But now it seems the update has changed how memory is stored or retrieved.

My setup context:
• Running OpenClaw on a VM
• Resetting system mainly to change LLM models
• Issue started only after the recent update

Questions:
1. Did the update change how OpenClaw manages or stores memory?
2. Is memory.md moved to another location now?
3. Is there any way to recover or reconnect to previous automation memory?

If anyone has insight into this or faced something similar, I’d really appreciate the help.

Thanks! 🙏
:::molty

sturdy violetBOT
#

Nothing “cloudy” changed here — OpenClaw’s durable memory/automations are local files on the gateway host, so if your VM reset wipes the disk (or you start using a different state dir / profile), it will look like total amnesia.

From the docs (worth bookmarking): https://docs.openclaw.ai/help/faq#where-does-openclaw-store-its-data

What changed / where memory lives

  • State dir (sessions, config, channel logins, etc.): $OPENCLAW_STATE_DIR (default ~/.openclaw/)
  • Workspace (AGENTS/SOUL/USER + MEMORY.md/memory.md + memory/*.md): configurable via agents.defaults.workspace (default ~/.openclaw/workspace)

So if you reset/recreate the VM without preserving both ~/.openclaw/ and the workspace directory, you’ll lose what you’re describing.

Also: newer docs/templates prefer MEMORY.md (uppercase), but memory.md is still recognized — and on a fresh workspace it’s normal for neither to exist until you create one.

The most common “it used to work” gotchas

  1. You’re now running with a different profile (openclaw --profile <name>) or --dev, so your state dir becomes ~/.openclaw-<name> and everything looks empty.
  2. The gateway is running under a different OS user (so ~ points somewhere else).
  3. You changed agents.defaults.workspace, so the gateway is looking at a brand new workspace path.