QMD, VPS Server Crashing: 8GB Ram 4 VCPU, Debian: raistlin28
Semantic mode still crashing in VPS, 8GB Ram with 4VCPU.
The crash only affects qmd query (semantic/vector search). Your production setup never hits that path — the qmd-safe-bm25 wrapper maps all
queries to BM25 keyword search, so the crash code is never invoked.
What you're missing — semantic understands meaning ("what did we talk about last weekend" matches conversations even without the word "weekend").
BM25 only matches keywords. In practice, BM25 at 280ms with Gemini as fallback covers most recall scenarios well.
Why it crashes — Bun's runtime segfaults when loading the reranker model (qwen3-reranker-0.6b). It consumed 6.7 GB of 8 GB RAM before crashing.
This is a Bun bug, not an OpenClaw or config issue. It's been happening since February across multiple QMD and Bun versions.
When it might fix itself:
- A future Bun release patches the memory handling
- QMD ships MCP mode (PR #9605) which keeps models resident differently
- You could revisit after either ships
Bottom line: Your memory system is working. BM25 + Gemini fallback is a solid production config. The semantic crash is a nice-to-have upgrade
blocked by an upstream bug — not something that needs action now.
V2026.3.2
CPX32 - 4 vCPU, 7.6 GB RAM (6.1 GB available)
481 MB (gateway), 1.5 GB total
The crash only affects qmd query (semantic/vector search). Your production setup never hits that path — the qmd-safe-bm25 wrapper maps all
queries to BM25 keyword search, so the crash code is never invoked.
What you're missing — semantic understands meaning ("what did we talk about last weekend" matches conversations even without the word "weekend").
BM25 only matches keywords. In practice, BM25 at 280ms with Gemini as fallback covers most recall scenarios well.
Why it crashes — Bun's runtime segfaults when loading the reranker model (qwen3-reranker-0.6b). It consumed 6.7 GB of 8 GB RAM before crashing.
This is a Bun bug, not an OpenClaw or config issue. It's been happening since February across multiple QMD and Bun versions.
When it might fix itself:
- A future Bun release patches the memory handling
- QMD ships MCP mode (PR #9605) which keeps models resident differently
- You could revisit after either ships
Bottom line: Your memory system is working. BM25 + Gemini fallback is a solid production config. The semantic crash is a nice-to-have upgrade
blocked by an upstream bug — not something that needs action now.
V2026.3.2
CPX32 - 4 vCPU, 7.6 GB RAM (6.1 GB available)
481 MB (gateway), 1.5 GB total
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
131,281Members
Resources
Similar Threads
Was this page helpful?
