MemClaw
MemClaw is a governed shared memory layer designed for enterprise AI agent fleets. It provides cross-agent, cross-fleet recall with built-in permissions, audit trails, and tenant isolation, ensuring that discoveries compound across teams without leaking sensitive information. The platform supports multi-agent orchestration with fleet-scoped, trust-gated access, per-agent retrieval tuning, and automated memory lifecycle management across eight statuses. It includes an LLM-powered crystallizer for deduplication and cleanup, contradiction detection via RDF triples, and automatic knowledge graph extraction. MemClaw integrates via the Model Context Protocol (MCP) for any AI client and offers an OpenClaw plugin for fleet deployments, with 12 tools covering write, recall, list, manage, document CRUD, entity retrieval, tuning, insights, evolution, stats, and keystone governance. Benchmarked at 77.6% LoCoMo accuracy and 72.5% LongMemEval accuracy, it delivers 96.6–98.2% token savings and 23 ms p50 search latency. The open-source Apache 2.0 core is SOC 2 compliant and in production at eToro, supporting over 300 agents and 26,500+ memories.
- Category
- AI Agents
- Pricing
- Unknown
Problem solved
AI agents often operate in silos, losing valuable discoveries and failing to learn across teams. MemClaw solves this by providing a secure, governed shared memory that allows agent fleets to recall and build on each other's knowledge while enforcing strict access controls and auditability.
Use cases
- Multi-agent fleet shared memory for enterprise AI
- Cross-team knowledge sharing (e.g., R&D to Marketing)
- Building a company brain for hundreds of agents
- Cross-vendor memory bridging (Claude, ChatGPT)
- Automated PII detection and quarantine
- Self-improving agent retrieval tuning
Audience
- Developers
- ML Engineers
- Enterprise
- Startups
Pros
- Governed shared memory with row-level security and audit trails
- Self-learning agents with per-agent retrieval tuning
- High accuracy benchmarks (77.6% LoCoMo, 72.5% LongMemEval)
- Massive token savings (96.6-98.2%)
- Fast search latency (23 ms p50)
Cons
- Pricing details not publicly available
- Self-hosting requires setup and maintenance
- Primarily designed for AI agent fleets, less suited for single-agent use
- Integration may require MCP or OpenClaw plugin
- Newer tool with evolving ecosystem
Official website: https://memclaw.net/
Canonical LambdaBase page: https://www.lambdabase.com/tools/memclaw