OpenClaw Memory

**Judgement:** A mature, well-designed core memory skill for OpenClaw that prioritizes transparency, durability, and control over automatic “black box” memory. Strong choice for technical users who want predictable, debuggable memory behavior and are comfortable managing configs and API keys. **What it does well** - Enforces a clean separation between **daily logs** and **curated long‑term memory**, both stored as Markdown on disk, making the agent’s memory easy to inspect, edit, and back up. - Provides robust **hybrid search** (vectors + BM25) with configurable backends, temporal decay, and MMR, giving high-quality recall even across long histories. - Mitigates context-window limits via an **automatic memory flush** before compaction, so important instructions and facts can be written to disk rather than silently lost. - Offers advanced options like **QMD** for local, higher-quality retrieval, plus `extraPaths` for using external Markdown knowledge bases. **Key risks / limitations** - **Configuration sensitivity:** Reliable semantic search depends on correctly setting up an embedding provider and API keys; misconfiguration leads to weak or nonexistent memory_search behavior. - **Operational complexity:** Power features (QMD backend, multiple providers, temporal decay/MMR tuning) are best suited to users comfortable editing JSON config and maintaining local tools. - **Privacy and data hygiene:** While memory is local Markdown, there’s no built‑in redaction or PII policy; users must ensure they don’t persist secrets or sensitive personal data where they shouldn’t, especially if syncing via git or shared storage. - **Session-type constraints:** MEMORY.md only loads in private/main sessions and large files are truncated; misuse or oversized files can create subtle “the agent forgot” scenarios if not monitored. **Recommended scenarios** - Primary memory system for developers or power users running OpenClaw locally who want **fine-grained control and visibility** over what the agent remembers. - Long-running projects where **decisions, rules, and project context** must persist and be reliably retrievable over time. - Setups where you want to **compose with other memory plugins** (Mem0, Cognee, Supermemory) but still maintain a strong, predictable baseline of local file-backed memory. - Less ideal for completely non-technical users or highly regulated settings unless paired with additional governance (e.g., clear redaction practices, storage policies).