memory-search

Memory Search

Lightweight parallel search across all memory sources. Cache hits bypass the full search.

Cache

Cache dir: ~/llm_wiki/.cache/memory-search/

• Lookup: Check cache first — if query-hash.json exists and TTL not expired, return cached results
• Write: After all sources return, write merged results to cache
• TTL: 1 hour default (override per-query if needed)
• Key: SHA-256 of canonicalized query (lowercased, stripped stop words)

Memory Sources

1. ~/roadmap — Project roadmaps and planning docs (~/roadmap/)
2. beads — Issue/bead tracking (~/.claude/projects/*/memory/*.md or .beads/issues.jsonl)
3. claude memories — Session memories (~/.claude/projects/*/memory/)
4. hermes sqlite — ~/.hermes/memory.db
5. hermes briefings — ~/.hermes/memory/briefing-*.md and mcp-mail-ack-log.md
6. hermes index — ~/.hermes/MEMORY.md
7. openclaw memories — ~/openclaw-repo/MEMORY.md, ~/.hermes/memory/
8. wiki — ~/llm_wiki/ (via wiki-search)
9. history — ~/.claude/projects/*/*.jsonl

Note: Mem0 (Qdrant at localhost:6333) not directly searchable — skip.

Execution

Run all searches in parallel via /e subagents:

/e Search ~/roadmap for "$QUERY". List files with matching snippets.

/e Search beads for "$QUERY" using: br search "$QUERY" --json 2>/dev/null | head -40. NEVER read .beads/issues.jsonl directly — it is 1MB+ and forbidden. Also check ~/.claude/projects/*/memory/*.md for matching bead IDs and titles.

/e Search ~/.claude/projects/*/memory/ for "$QUERY". Search MEMORY.md indexes and individual .md files. Show snippets.

/e Search ~/.hermes/memory.db: sqlite3 ~/.hermes/memory.db "SELECT timestamp, source, content FROM memories WHERE content LIKE '%$QUERY%' LIMIT 20;"

/e Search ~/.hermes/memory/briefing-*.md and ~/.hermes/memory/mcp-mail-ack-log.md for "$QUERY" using: grep -m 5 -n "$QUERY" ~/.hermes/memory/briefing-*.md ~/.hermes/memory/mcp-mail-ack-log.md 2>/dev/null | head -20. The -m 5 per-file limit prevents reading full large files.

/e Search ~/.hermes/MEMORY.md for "$QUERY". Show matching entries.

/e Search ~/openclaw-repo/MEMORY.md and ~/.hermes/memory/ for "$QUERY". Show snippets.

/wiki-search $QUERY

/e Search ~/.claude/projects/*/*.jsonl for "$QUERY" using two-phase partial read: (1) grep -rl "$QUERY" ~/.claude/projects/*/*.jsonl 2>/dev/null | head -5 to get matching filenames without reading content; (2) for each file: grep -m 2 "$QUERY" <file> 2>/dev/null | cut -c1-200 to get up to 2 matching lines, truncated to 200 chars each. Full command: grep -rl "$QUERY" ~/.claude/projects/*/*.jsonl 2>/dev/null | head -5 | xargs -I{} grep -m 2 "$QUERY" {} 2>/dev/null | cut -c1-200 | head -20. The -m 2 flag stops grep after 2 matches per file so it never reads to EOF. DO NOT use grep -H without -m (reads full file).

Aggregation

Wait for all 9 subagents. Merge results into sections:

# Memory Search: "$QUERY"

## ~/roadmap
[results]

## Beads
[results]

## Claude Memories
[results]

## Hermes SQLite
[results]

## Hermes Briefings
[results]

## Hermes Index
[results]

## OpenClaw
[results]

## Wiki
[results]

## History
[results]

If a source returns nothing, mark as "— no matches". Sort by relevance within each section.