Summary

memory_search can block the foreground tool call for minutes when the first search returns zero results, because it synchronously awaits manager.sync({ reason: "search", force: true }) before returning to the agent.

This makes a normal "no results" search miss behave like a full memory indexing/sync operation inside the user-facing tool call deadline.

Observed Behavior

On OpenClaw 2026.6.11, a direct dynamic memory_search call for a query with no hits repeatedly timed out in the foreground tool layer:

memory_search timed out after 15s
OpenClaw dynamic tool call timed out after 90000ms while running tool memory_search

Queries with existing hits returned normally once the miss path was avoided.

After locally disabling the foreground sync-on-miss branch in the generated dist/tools-*.js, both cases returned immediately:

memory_search("warmup") -> [] in ~28ms
memory_search("fruit juice farm noon appmagic") -> 3 results in ~17ms

Root Cause

The memory_search tool currently performs a synchronous full sync when rawResults.length === 0:

if (rawResults.length === 0 && activeMemory.manager.sync) {
  await activeMemory.manager.sync({
    reason: "search",
    force: true
  });
  rawResults = await activeMemory.manager.search(query, searchOptions);
}

For local QMD-backed memory stores, sync may involve filesystem scanning, vector updates, embedding calls, local model initialization, or other expensive work. That work should not block a foreground agent tool call, especially under 15s/90s watchdogs.

Expected Behavior

A search miss should return an empty result set promptly.

If OpenClaw wants to improve freshness after a miss, it should schedule a background sync that does not block the user-facing tool call.

Suggested Fix

Add a config flag, defaulting to non-blocking behavior:

{
  "memorySearch": {
    "syncOnMiss": false
  }
}

Possible implementation:

if (rawResults.length === 0 && activeMemory.manager.sync && syncOnMiss) {
  if (syncOnMiss === "background") {
    activeMemory.manager.sync({ reason: "search", force: true }).catch((err) => {
      log.warn({ err }, "memory_search background sync-on-miss failed");
    });
  } else {
    await activeMemory.manager.sync({ reason: "search", force: true });
    rawResults = await activeMemory.manager.search(query, searchOptions);
  }
}

Recommended default: false or "background", not foreground await.

Why This Matters

Agent tool calls are latency-sensitive. A no-hit memory search is a normal outcome and should not trigger an expensive synchronous maintenance path. In production assistants, this can make memory appear broken even though the index and search backend are healthy.