Memory plugin for OpenClaw that runs 100% locally. Smarter search, memory quality checks, and compaction rescue — so your agent remembers everything, no matter how long the conversation. No API keys. No subscriptions. Just install and go.
$49 — one-time purchase, no subscriptionThree features that make long conversations actually work.
Native memory does flat text search. Memory Stack builds a local knowledge graph from your conversations — tracking projects, people, APIs, and how they connect. Ask "what depends on service X?" and get real answers.
Memories pile up and get noisy. Memory Stack finds clusters of similar entries and suggests consolidation — turning 5 scattered notes about the same topic into one clear summary. Your memory gets sharper over time, not messier.
Long conversations get compressed — and key decisions get lost. Memory Stack saves important facts (decisions, deadlines, requirements) before compression happens, and brings them back automatically. No matter how long the conversation.
MMR diversity ranking prevents redundant results. Temporal decay prioritizes recent memories. Session indexing searches across past conversations. All running locally — no API keys, no extra cost.
Built-in quality checks find duplicates, stale entries, and noise. Scores your memory 0-100 and tells you exactly what to clean up. Never deletes anything automatically — you stay in control.
Install, restart OpenClaw, done. No embedding providers to configure, no API keys to manage, no project initialization needed. Memory improves the moment you install it.
Memory Stack plugs into OpenClaw as a native memory provider. No manual setup after install.
OpenClaw's native memory is good. Memory Stack makes it better.
| Feature | Native Memory | Memory Stack |
|---|---|---|
| Text search (BM25) | Yes | Yes |
| Vector search | Yes (needs API key or local setup) | Yes (QMD, zero config) |
| Hybrid search + MMR | Yes | Yes |
| Temporal decay | Yes | Yes |
| Knowledge graph | No | Yes — tracks entities and relationships |
| Self-evolving memory | No | Yes — finds duplicates, suggests consolidation |
| Compaction rescue | LLM-based flush only (costs tokens) | Rule-based extraction (zero token cost) |
| Memory health check | No | Yes — scores quality 0-100 |
| Multi-engine search | Single backend | 5 engines with fallback routing |
| Session search | Experimental | Yes |
| Embedding setup needed | Yes — requires provider config | No — works out of the box |