ClawMem
ClawMem is a local-first context engine/memory for AI coding agents. It indexes markdown/project documents into a SQLite-based “vault”, performs hybrid retrieval (BM25 + vector + reranking + graph/intent-based traversal), and integrates with Claude Code via hooks + an MCP stdio server and with OpenClaw via a ContextEngine plugin or MCP tools. It also captures session decisions/handoffs and writes them back to the same local vault; it includes measures against prompt injection in surfaced content and supports deduplication, lifecycle (pin/snooze/archive), and crash-safe transaction-safe indexing.
Score Breakdown
⚙ Agent Friendliness
🔒 Security
Local-first design with no API keys/cloud dependencies reduces exposure. However, no explicit authN/authZ is described for MCP/local tools; therefore security relies on local host trust and vault filesystem permissions. The README claims “guards against prompt injection” for surfaced content and mentions transaction-safe indexing and crash-safe behavior, which can reduce certain integrity risks. Dependency hygiene is not verifiable from provided manifest/README; scores are estimated based on explicit dependencies without CVE review.
⚡ Reliability
Best When
You want an on-device, agent-integrated memory layer that can surface relevant context automatically and also provide tool-based retrieval (MCP/OpenClaw) while keeping data local in a SQLite vault.
Avoid When
You need a centralized, network-accessible API or require robust multi-user authentication/authorization at the service layer.
Use Cases
- • Persistent coding agent memory for Claude Code and OpenClaw across sessions/projects
- • Retrieval-augmented grounding over local markdown notes, docs, and research dumps
- • Session decision capture and handoff generation for long-running coding workflows
- • Targeted retrieval via MCP tools (timeline, causal links, intent search) when hooks don’t surface enough context
- • Local-only “hybrid RAG + memory graph” for privacy-sensitive developer teams
Not For
- • Use as a hosted/cloud service for remote memory (it is explicitly local/no cloud dependencies)
- • Use in environments where Bun/system dependencies cannot be installed
- • Multi-tenant server deployments without strong additional isolation controls (primarily designed for local single-user vaults)
- • Environments needing a public REST/GraphQL API for external systems (no such interface is described)
Interface
Authentication
No authentication mechanism for MCP or local tools is described; the primary security model appears to be local execution and filesystem/vault isolation.
Pricing
Open-source (MIT). Costs are primarily compute/storage for local indexing and optional model servers (llama.cpp), not service pricing.
Agent Metadata
Known Gotchas
- ⚠ Designed as a local system; agents must run where the vault/models exist.
- ⚠ No authentication for MCP tools is described; in practice, ensure only trusted local clients can connect.
- ⚠ Integration modes share the same SQLite vault—concurrent access exists (WAL + busy_timeout mentioned) but agent workloads that heavily parallelize may still contend.
- ⚠ “Hooks handle 90%” suggests relying purely on MCP may miss lifecycle orchestration such as compaction and afterTurn/feedback-loop flows unless using the OpenClaw ContextEngine plugin.
Alternatives
Full Evaluation Report
Comprehensive deep-dive: security analysis, reliability audit, agent experience review, cost modeling, competitive positioning, and improvement roadmap for ClawMem.
AI-powered analysis · PDF + markdown · Delivered within 30 minutes
Package Brief
Quick verdict, integration guide, cost projections, gotchas with workarounds, and alternatives comparison.
Delivered within 10 minutes
Score Monitoring
Get alerted when this package's AF, security, or reliability scores change significantly. Stay ahead of regressions.
Continuous monitoring
Scores are editorial opinions as of 2026-03-30.