Summary
OpenClaw memory_search API returns 0 hits for specific terms despite content being verified present and indexed.
Environment
- OpenClaw version: 2026.3.2 (85377a2)
- OS: macOS 26.3.0 (arm64)
- Provider: openai/text-embedding-3-small
- Agent: tex
- Index status: 7/7 files, 45 chunks indexed
Issue Details
Query 'Rocky Point bagging' returns 0 hits despite 17+ instances verified across indexed files via grep.
Affected queries (all return 0 hits):
- 'Rocky Point bagging'
- 'bagging operations'
- 'core business unit'
- 'retail garden products'
Working queries on same index:
- 'Norwell Valley' (2 hits)
- 'Waree portfolio' (4 hits)
- 'Josh Keith' (1 hit)
Reproduction Steps
- Index files containing target terms
- Run memory_search with specific terms
- Observe 0 hits despite content presence
Remediation Attempted
- Database recreation (3x with rm ~/.openclaw/memory/tex.sqlite)
- Forced reindexing (openclaw memory index --agent tex --force)
- Content modifications and alternative phrasings
- Embedding cache clearing
- All unsuccessful
Expected Behavior
memory_search should return hits for terms present in indexed content
Actual Behavior
Specific terms consistently return 0 hits while other queries on same index work normally
Workaround
CLI search works: openclaw memory search --agent tex --query 'Rocky Point bagging'
Impact
Strategic knowledge retrieval may fail unpredictably for business-critical terms
Summary
OpenClaw memory_search API returns 0 hits for specific terms despite content being verified present and indexed.
Environment
Issue Details
Query 'Rocky Point bagging' returns 0 hits despite 17+ instances verified across indexed files via grep.
Affected queries (all return 0 hits):
Working queries on same index:
Reproduction Steps
Remediation Attempted
Expected Behavior
memory_search should return hits for terms present in indexed content
Actual Behavior
Specific terms consistently return 0 hits while other queries on same index work normally
Workaround
CLI search works: openclaw memory search --agent tex --query 'Rocky Point bagging'
Impact
Strategic knowledge retrieval may fail unpredictably for business-critical terms