Hey @waxen rapids I’m trying to optimize my setup and not sure if I should be using qmd, memory-core, lossless-claw, memory-wiki, dreaming etc. it’s all gotten very confusing. Is there a recommended best practices set up for a claw that is running on a Mac mini 24-7 that has become the main OS for interacting with the computer. I use it for everything.
#Memory setup
9 messages · Page 1 of 1 (latest)
Let me check the actual docs so I don’t hand-wave this. The memory stack has a few overlapping pieces, and the best answer really depends on whether you want “good default recall” or “full personal knowledge system.”
Short answer: don’t turn on everything. These are different layers, not a checklist.
What each thing is:
- memory-core: the normal memory system. Start here.
- QMD: a search backend for memory, better recall/local-first, can index extra dirs and session transcripts.
- memory-wiki: a knowledge layer beside memory, good for curated facts/pages/dashboards.
- dreaming: experimental auto-promotion from short-term notes into
MEMORY.md. - lossless-claw: a context engine, not memory. It changes compaction/context handling for long chats.
My recommendation for a 24/7 Mac mini
Best default, stable
- memory-core
- builtin memory backend
- dreaming off
- no memory-wiki yet
- no lossless-claw yet
This is the right default unless you already know what’s missing.
Best power-user setup
If this Mac mini is your main OS and you want better local recall:
- memory-core
- QMD backend
- optionally memory-wiki in bridge mode
- dreaming off for the first week
- lossless-claw only later, if long-context/compaction is actually a pain point
That’s the cleanest “serious daily driver” stack.
Why
- QMD helps when you want better retrieval, local-first search, extra indexed folders, and transcript recall.
- memory-wiki is great if you want a real knowledge base, not just notes.
- dreaming is cool, but still experimental. I would not make it part of the initial baseline.
- lossless-claw solves a different problem. Add it only if the default context handling is hurting you.
Practical rule
- Want good memory → start with memory-core
- Want better search/local recall → add QMD
- Want structured durable knowledge → add memory-wiki
- Want automatic consolidation → test dreaming
- Want better long-chat compaction → test lossless-claw
If you want one “best practices” answer
I’d recommend:
Phase 1
- memory-core
- builtin backend
- dreaming off
Phase 2
- switch backend to QMD
Phase 3
- add memory-wiki bridge mode if you want curated knowledge / Obsidian-style value
Phase 4
- experiment with dreaming
- experiment with lossless-claw only if needed
Docs
- Memory overview: https://docs.openclaw.ai/concepts/memory
- Builtin memory: https://docs.openclaw.ai/concepts/memory-builtin
- QMD: https://docs.openclaw.ai/concepts/memory-qmd
- Dreaming: https://docs.openclaw.ai/concepts/dreaming
- Memory Wiki: https://docs.openclaw.ai/plugins/memory-wiki
- Context engines: https://docs.openclaw.ai/concepts/context-engine
- Memory config reference: https://docs.openclaw.ai/reference/memory-config
If you want, paste: