openclaw - 💡(How to fix) Fix feat(active-memory): make recall budget split configurable (maxMemoryResults, snippet length, wiki max chars) [1 comments, 2 participants]

Problem

The active-memory recall budget (maxSummaryChars, default 1000) is shared between memory search and wiki (vectordb) search results. Currently the allocation is hardcoded in the recall function:

• maxMemoryResults = 5 × 400 char snippets → memory consumes ~997 of 1000 chars
• wikiTopK = 3 × 2500 char budget → gets truncated to almost zero because memory eats the budget first
• wikiMaxChars = 2500 (internal budget for wiki formatting)

Result: Wiki search results are effectively never injected into the prompt because memory fills the entire 1000-char summary budget.

Reproduction

1. Configure maxSummaryChars: 1000 (default)
2. Have both memory and wiki search return results
3. Observe that only memory snippets appear in the <active_memory_plugin> block
4. Wiki search returns valid hits (verified independently) but they never make it into the injected summary

Proposed Fix

Move the hardcoded values into plugin config:

{
  "plugins": {
    "entries": {
      "active-memory": {
        "config": {
          "maxMemoryResults": 3,
          "memorySnippetChars": 200,
          "wikiTopK": 3,
          "wikiMaxChars": 2500
        }
      }
    }
  }
}

Or alternatively: implement a proportional budget split (e.g., 50/50 between memory and wiki) instead of first-come-first-served.

Impact

• Users with active-memory + wikis integration lose wiki context injection entirely
• Hardcoded values in compiled JS are fragile (local patches get overwritten on update)
• No way to tune the trade-off between memory recall depth and wiki coverage

Environment

• OpenClaw 2026.4.29 (a448042)
• active-memory plugin with local vectordb.py integration
• maxSummaryChars: 1000 (default, enforced by latest update)