Auto-recall hardcodes maxResults: 10, ignoring recall.maxResultsDefault config

[Z710210920]

Problem

The auto-recall feature in the before_prompt_build hook currently hardcodes maxResults: 10 for both local and Hub memory searches:

https://github.com/MemTensor/MemOS/blob/main/apps/memos-local-openclaw/index.ts#L1893-L1896

This means the recall.maxResultsDefault configuration option is effectively ignored during automatic recall, even though users may set it to a lower value (e.g., 3 or 5) to reduce token consumption.

Current Behavior

• memory_search tool correctly respects maxResultsDefault (default 6)
• Auto-recall always fetches 10 results from local + 10 from Hub, regardless of config
• After LLM filtering and deduplication, this still injects ~5-10 memory entries into every system prompt
• At ~300-500 tokens per entry, this adds ~2,000-5,000 tokens per turn

Expected Behavior

Auto-recall should either:

1. Read from recall.maxResultsDefault config instead of hardcoding 10, OR
2. Support a separate recall.autoRecallMaxResults config key for independent control

Suggested Fix

// index.ts ~1893
const recallCfg = ctx.config.recall ?? {};
const autoRecallMax = recallCfg.maxResultsDefault ?? 10;

const arLocalP = engine.search({
  query,
  maxResults: autoRecallMax,
  minScore: 0.45,
  ownerFilter: recallOwnerFilter
});

Environment

• Plugin version: 1.0.8
• OpenClaw version: 2026.4.x

Thank you for the excellent plugin!

[MaiHHConnect]

💬 我也在做记忆系统相关的工作：CausaMem，可以来看看，说不定有能借鉴的地方 🙂