#2456
Issues/milla-jovovich/mempalace/#2456

memplace stucks

Openalifiroozi80opened 1mo ago

Hey

mempalace init . --yes

stuck during

MemPalace Mine

phase. it was working just fine and from one day, randomly it happend. now it being timeout if agents calls it. I tried upgrading it, I also removed and reinstalled it, but no luck.

mempalace init . --yes
  No LLM provider reachable (Cannot reach Ollama at http://localhost:11434: <urlopen error [Errno 61] Connection refused>). Running heuristics-only — pass --no-llm to silence this.
  Corpus origin: not AI-dialogue (confidence: 0.90)

  Scanning for entities in: .

==========================================================
  MemPalace -- Entity Detection
==========================================================

  Scanned your files. Here's what we found:


  PEOPLE:
     1. XXX          [####.] 655 commits across 1 repo

  PROJECTS:
     1. xxx         [###..] 1000 commits (none by you)

  Auto-accepting 4 people, 1 projects, 0 topics.
  Entities saved: /Users/xxx/Desktop/pro/xxx/xxx/entities.json
  Registry updated: /Users/xxx/.mempalace/known_entities.json

=======================================================
  MemPalace Init -- Local setup
=======================================================

  WING: XXX
  (653 files found, rooms detected from folder structure)

    ROOM: XXX
          Files from XXX
    ...

-------------------------------------------------------

  Config saved: /Users/xxx/Desktop/pro/xxx/availability/xxx/mempalace.yaml

  Next step:
    mempalace mine .

=======================================================

  ~653 files (~2 MB) would be mined into this palace.

  Mine this directory now? [Y/n] Y

=======================================================
  MemPalace Mine
=======================================================
  Wing:    xxx
  Rooms:   xxx ...
  Files:   653
  Palace:  /Users/xxx/.mempalace/palace
  Device:  coreml
-------------------------------------------------------

Environment:

  • OS: MacOS M5 Tahoe (26.6.2)
  • Python version: 3.14.7
  • MemPal version: 3.9.0

Activity

5
igorls1mo ago

Triaging: the symptom (macOS + Device: coreml, mine hangs at start, upgrade/reinstall doesn't help) looks like the same root cause as #2224 — CoreML provider stalling on the small quantized graph with CPU spikes. If you set MEMPALACE_EMBEDDING_DEVICE=cpu (or unset it) and the mine completes, that confirms it and we can close this in favour of #2224. Also worth ruling out #2132 (unbounded entity extraction spinning) — the trace stops right after the entity/rooms banner, so if it hangs there even on CPU, that is the other suspect.

alifiroozi801mo ago

Hey @igorls Thanks, how to set that?

set MEMPALACE_EMBEDDING_DEVICE=cpu

But still, in memplace init, I see Device: coreml.

igorls1mo ago

set is Windows cmd syntax — on macOS it doesn't export anything, so the process never sees it. Use one of these instead:

# one-off, just for this command
MEMPALACE_EMBEDDING_DEVICE=cpu mempalace init . --yes

# or export for the whole shell session
export MEMPALACE_EMBEDDING_DEVICE=cpu
mempalace init . --yes

If you want it to stick without env vars, add it to ~/.mempalace/config.json:

{ "embedding_device": "cpu" }

Either way the mine banner should then show Device: cpu. If the mine completes on CPU, that confirms this is #2224 (CoreML stalling); if it still hangs right after the entity/rooms banner even on CPU, that points to #2132 instead — please paste which one you see.

alifiroozi801mo ago

Yes, it still hangs even when it shows Device: cpu

alifiroozi801mo ago

For future searchers: I switched to ai-memory. Until now, it has worked just fine.