47a8b40fdb
TREE(74): training modules, entity profiles, 2890 references, keyword indices SPROUT(42): knowledge pages, project docs, curated source material SEED(164): daily notes, raw session logs, unprocessed material Updated AUDIT_MANIFEST.json with growth classifications.
3.1 KiB
3.1 KiB
pageType, id, title, sourceType, sourcePath, ingestedAt, updatedAt, status, growth
| pageType | id | title | sourceType | sourcePath | ingestedAt | updatedAt | status | growth |
|---|---|---|---|---|---|---|---|---|
| source | source.psb-thinking-2026-04-23 | psb-thinking-2026-04-23 | local-file | /home/topher/.openclaw/workspace-psb-thinking/memory/2026-04-23.md | 2026-05-03T01:58:16.317Z | 2026-05-03T01:58:16.317Z | active | seed |
psb-thinking-2026-04-23
Source
- Type:
local-file - Path:
/home/topher/.openclaw/workspace-psb-thinking/memory/2026-04-23.md - Bytes: 2464
- Updated: 2026-05-03T01:58:16.317Z
Content
# Memory - 2026-04-23
## SQL Database Question
- Topher asked if we're using SQL database — confusion from Crash bot
- **Answer:** No SQL — we use Qdrant (vector DB for RAG). SQLite IS used but only for OpenClaw's internal state (memory/*.sqlite, flows/registry.sqlite, tasks/runs.sqlite)
- Crash bot was talking about OpenClaw's own internal SQLite dbs — those are normal, not something we set up
## Memory Search Investigation (Key Finding)
- Topher questioned whether memory search is actually working via Ollama
- **Discovery:** Config says Ollama/nomic-embed-text, but ACTUAL backend is GGUF model via node-llama-cpp CPU
- Evidence: Every memory_search result in transcripts shows `hf:ggml-org/embeddinggemma-300m-qat-q8_0-GGUF`
- CLI test: `openclaw memory search` took 24 seconds (node-llama-cpp CPU fallback, no GPU)
- Ollama IS running (localhost:11434) but NOT being used for embeddings — only for chat
- **Root cause:** node-llama-cpp GGUF model exists at ~/.node-llama-cpp/models/ — "local" provider picks this by default, bypassing Ollama config
- node-llama-cpp has prebuilt Vulkan binary but system doesn't support it → falls back to CPU
## OpenClaw Memory Research Findings
Sources: velvetshark.com, docs.openclaw.ai, GitHub issues, community guides
### Three Failure Modes
1. **Failure A:** Never stored — instruction only in chat, never written to file
2. **Failure B:** Compaction destroyed it — lossy summary drops details/nuance
3. **Failure C:** Session pruning trimmed tool results (temporary)
### Known Bugs (OpenClaw v2026.4.8)
- **#57410:** Compaction results in full reset instead of compression (context 700K→29K, Compactions:0)
- **#17727:** Post-compaction, agent loses awareness of AGENTS.md/SOUL.md (re-injected files not re-injected after summarize)
- **#56072:** Daily 4AM reset silently archives session WITHOUT memory flush — context lost silently
### Missing Config
Our openclaw.json has NO compaction tuning:
- No `reserveTokensFloor` (defaults to 20K — too tight)
- No `memoryFlush` explicit config
- No `contextPruning` config
- No `compaction.notifyUser`
### What We Have
- memorySearch: enabled ✓
- Bootstrap files: 24K chars total — well under 150K cap ✓
- WAL on: registry.sqlite, runs.sqlite only (system dbs)
- Agent memory dbs: all delete journal mode
## WAL Protocol
- WAL enforcement confirmed running (HEARTBEAT.md checklist active)
- Updated SESSION-STATE.md with new findings
Notes
Related
- No related pages yet.