Skip to content

Core vs Advanced Labs

openclaw-mem has one public core and several optional advanced lanes.

If you are evaluating the product for the first time, start with Core. Advanced Labs are useful only after you trust the Store / Pack / Observe loop.

Core product

Surface Status Why it belongs in Core
Store Core Capture or ingest local memory records into inspectable JSONL/SQLite artifacts.
Pack Core Produce bounded ContextPack output with citations, trust policy, and trace receipts.
Observe Core Inspect timelines, exact records, pack traces, receipts, and rollback-friendly artifacts.
Sidecar install Core Lets operators evaluate without replacing the active OpenClaw memory backend.
Trust-policy synthetic proof Core proof Shows, on public synthetic data, how a trust-aware pack excludes quarantined content with an explicit reason.
Optional mem engine promotion Core adoption path Extends the same Pack contract into live-turn use only when the operator opts in.

Advanced Labs

These lanes are intentionally not required for the first evaluation path.

Surface Status Keep-out tape
Graph routing / synthesis Advanced Lab Useful for topology-aware recall experiments; not required for basic Store / Pack / Observe.
GBrain sidecar Experimental Lab Read-only lookup and restricted helper-job experiments; not a second truth store.
Continuity / self-model side-car Advanced Lab Derived continuity inspection and public-safe summaries; not a memory source of truth.
Dream Lite / dreaming director Experimental Lab Research-grade suggestion and rehearsal workflows; not part of the default memory path.
Optimize assist / autonomy loops Advanced Lab Governed maintenance workflows for mature operators; not needed to prove the core wedge.
Self Curator engine Advanced Lab Checkpointed lifecycle review/apply loops. Current scheduled lane may mutate SKILL.md body sections with rollback receipts; memory/dream/authority surfaces remain gated.
Command-aware compaction Advanced Lab Observe-path artifact handling for long outputs; not required for the first proof.

Evaluation rule of thumb

If a feature does not help you answer one of these questions, postpone it:

  1. Can I store local memory records in a way I can inspect?
  2. Can I pack relevant context with citations and receipts?
  3. Can I explain why a memory was included or excluded?
  4. Can I adopt this beside my current OpenClaw memory without a forced backend swap?

When those are clear, Advanced Labs are available for deeper operations.