Skip to content

Document practical requirements for QMD memory backend on CPU-only VPS deployments #90746

Description

@holgergruenhagen

Document practical requirements for QMD memory backend on CPU-only VPS deployments

QMD can be used as a local memory backend in OpenClaw, which is a very useful option for privacy and local control.

This issue is specifically about documenting when QMD is practical in OpenClaw deployments that run on CPU-only VPS infrastructure.

Related QMD upstream issue: tobi/qmd#709

Environment observed

OpenClaw was running with QMD on the following VPS:

  • Provider: Hetzner
  • Instance type: CCX23
  • Cost: about 30 USD per month
  • CPU: 4 dedicated vCPU cores
  • GPU: none
  • Use case: QMD as local memory backend for OpenClaw agents

Observed behavior

When QMD embedding runs on this machine, it can consume effectively all available CPU capacity.

On this 4-core dedicated CPU VPS, the embedding process was observed using around 300-400% CPU in top / htop, which means it was using nearly all available CPU cores.

This can make the machine much less responsive while embedding is running.

Why this should be documented in OpenClaw

The core question for OpenClaw users is not only how to limit CPU usage.

The bigger operational question is:

Is QMD a practical local memory backend for OpenClaw on CPU-only VPS deployments?

If yes, OpenClaw should document the minimum practical requirements and recommended settings.

If no, OpenClaw should make that clear, so users do not choose QMD for an environment where it is not expected to work well.

Questions to answer in the documentation

  1. Is QMD recommended as an OpenClaw memory backend on CPU-only VPS deployments?

  2. What are the minimum practical CPU requirements for using QMD with OpenClaw?

  3. Is a 4 dedicated core VPS, such as a Hetzner CCX23, sufficient for regular OpenClaw + QMD embedding work?

  4. What QMD settings should OpenClaw recommend for this exact kind of CPU-only deployment?

  5. Should OpenClaw ship safer default settings for QMD embedding jobs on machines without GPU acceleration?

  6. Should OpenClaw warn users when QMD is configured on a CPU-only host without explicit embedding limits?

Documentation improvement requested

It would be helpful if the OpenClaw documentation included a short section for QMD deployments that explains:

  • whether CPU-only QMD usage is recommended for OpenClaw
  • minimum practical CPU requirements
  • whether a 4 dedicated core VPS is enough
  • recommended QMD embedding settings for CPU-only operation
  • how OpenClaw users should configure scheduled embedding jobs safely
  • how QMD local embeddings differ from OpenClaw's separate memory_search / OpenAI embedding path

Notes

This issue is not about blaming QMD for using CPU. Local embedding is expected to need real compute.

The goal is to make the OpenClaw deployment path clearer, especially for users who run agents on rented VPS infrastructure and need to understand whether QMD is a good fit before enabling it.

Metadata

Metadata

Assignees

No one assigned

    Labels

    P3Low-priority cleanup, docs, polish, ergonomics, or speculative work.clawsweeper:fix-shape-clearClawSweeper found a clear likely implementation shape for this issue.clawsweeper:needs-maintainer-reviewClawSweeper marked this issue as needing maintainer review before automation.clawsweeper:needs-product-decisionClawSweeper marked this issue as needing a product or behavior decision.clawsweeper:no-new-fix-prClawSweeper does not recommend queueing a new automated fix PR for this issue.impact:otherThis issue has meaningful maintainer-visible impact outside the owned taxonomy.issue-rating: 🌊 off-meta tidepoolIssue quality rating does not apply to this item.

    Type

    No type

    Fields

    Priority

    None yet

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions