openclaw - 💡(How to fix) Fix [Feature]: expose QMD no-rerank for memory.qmd query mode [1 participants]

Summary

Expose QMD 2.1.0's no-rerank capability from OpenClaw's QMD memory backend so users can run searchMode: "query" without the reranking step.

Problem to solve

OpenClaw already supports:

• memory.backend = "qmd"
• memory.qmd.searchMode = "query" | "search" | "vsearch"

But when searchMode: "query" is used, there is currently no config or tool-level way to disable reranking.

Upstream QMD 2.1.0 added:

• CLI: qmd query --no-rerank
• MCP query tool: rerank: false

So OpenClaw currently cannot expose a useful middle ground between:

• search = fast but BM25-only
• query = best recall but slower
• desired = query without rerank, for faster hybrid results

Proposed solution

Expose the rerank toggle in one or both of these forms.

Option A: config-level

{
  "memory": {
    "backend": "qmd",
    "qmd": {
      "searchMode": "query",
      "noRerank": true
    }
  }
}

Expected behavior:

• ignore when searchMode !== "query"
• in direct CLI mode, append --no-rerank
• in MCP-backed query mode, pass rerank: false if supported

Option B: request-level

Expose a request-level toggle on memory_search, for example:

{
  "query": "QMD 2.1.0 rerank false",
  "maxResults": 5,
  "rerank": false
}

This would be more flexible than config-only support.

Alternatives considered

• Use memory.qmd.searchMode = "search": faster, but loses hybrid query behavior
• Keep using searchMode = "query": better recall, but slower than necessary when reranking is not needed
• Call qmd query --no-rerank manually outside OpenClaw: workable as a local workaround, but bypasses the normal memory_search path

Impact

Affected: users running memory.backend = "qmd", especially on CPU-only or latency-sensitive systems
Severity: Medium
Frequency: any interactive use of searchMode: "query"
Consequence: users must choose between BM25-only search and full reranked query mode, with no supported middle ground even though upstream QMD now provides one

Evidence/examples

QMD 2.1.0 release notes: https://github.com/tobi/qmd/releases/tag/v2.1.0

Relevant excerpt:

--no-rerank flag skips the reranking step in qmd query
useful when you want fast results or don't have a GPU
also exposed as rerank: false on the MCP query tool

In local verification, qmd query --no-rerank worked as expected and provided a useful speed/quality tradeoff compared with full reranked query mode.

Additional information

This is separate from warm-process / MCP-caching work. Even if QMD is kept warm through MCP, a rerank toggle is still useful as an explicit latency vs quality control.