Repository Issue Activity (beta)

#706 multi-get does not resolve docids despite docs/examples

#757 Quoted phrase queries containing dotted tokens return no results (e.g. "1.0.21"), the phrase-path half of #563

#640 Installing qmd with CUDA on WSL2

#556 qmd embed exit with "fetch failed"

#751 qmd mcp (stdio) doesn't exit on stdin EOF — orphans to PID 1 and leaks RAM after parent dies

#735 qmd embed deadlocks at 0% CPU on Apple M5 Max + macOS 26.4 — Metal shader JIT compile failure in bundled llama.cpp b8390 (distinct from #673/#724)

#747 Qwen3 reranker scores compressed (downstream of node-llama-cpp rankAll double-sigmoid)

#710 Overlapping qmd processes crash with SQLITE_BUSY (no busy_timeout): update dies at insertContent while an embed runs

#736 Store-open CJK FTS migration OOMs large indexes and can leave BM25 FTS empty

#741 Docs: mcp-setup.md still names search tool `structured_search` (renamed to `query`)

#743 Feature Request: Allow host configuration for qmd mcp --http (Docker support)

#717 Lossy path normalization breaks QMD_EDITOR_URI for files with spaces

#739 GPU selection falls back to CPU instead of Vulkan on non-NVIDIA machines

#738 `collection add`: trailing backslash in a quoted path swallows following flags — collection is created broken (0 files, wrong name/mask) but still reports success

#699 npm install fails (macOS apple silicon, latest v26)

#645 exclude: patterns in index.yml to prevent double-indexing nested collections

#724 qmd embed cannot complete large collections in one run — session maxDuration (10 min) aborts mid-run with mass "LLM session expired before embedding chunk" failures

#723 Nix flake bypasses `bin/qmd` launcher — silent Metal teardown noise on darwin + MCP stdio risk

#722 Nix package omits `skills/` — `qmd skill show` and `qmd skills list` always fail

#721 Alternative Qwen3 Rerankers don't work

#45 Lossy paths

#671 Scope `qmd update` to a specific collection with -c

#716 Per-subcommand --help and bench discoverability guardrails

#714 Qwen3-Embedding models return ~0% retrieval recall — pooling (last-token) not set, prompt formatting alone isn't enough

#709 Clarify whether QMD is practical for OpenClaw on CPU-only VPS instances