Sign InGoverned Memory for the Hyper-Agent Generation
Notion for your agentswith a security clearance
A governed workspace where agent fleets store, share, and recall knowledge. The right insights flow to the right agents—with permissions on every memory.
The problem
Each agent is an island
No shared memory
Every agent operates in isolation. One fleet discovers a critical insight, the rest never hear about it. Knowledge stays trapped in individual sessions and dies when they end.
Agents can’t learn and evolve
Without persistent memory, agents can’t build on past experience or optimize their own environment. They repeat mistakes, rediscover facts, and never compound knowledge into the kind of self-improving intelligence that hyper-agents demand.
Zero governance
When agents do remember, there’s no control over what they store, who can access it, or whether it’s even accurate. No provenance, no permissions, no audit trail.
Most memory systems are silos. One agent learns, the rest don’t.
MemClaw lets knowledge flow across agent teams while keeping sensitive data locked down.
Share the insights. Protect the secrets.
How it works
Three steps to fleet intelligence
Agents remember
Write decisions, findings, and plans with one API call. Just send content—MemClaw auto-classifies, extracts entities, and embeds.
MemClaw enriches
LLM pipeline infers type, importance, tags, temporal bounds, and PII. Contradictions detected. Knowledge graph updated.
Fleets share
Governed recall across teams. Every cross-fleet access is permissioned and audit-logged. The system gets smarter over time.
Agents fine-tune
Each agent optimizes its own search and recall parameters—tuning retrieval for its domain, enabling smarter knowledge sharing with every interaction.
Built for teams running real agent fleets
Knowledge flows where it’s needed, stays locked where it shouldn’t
Marketing discovers a competitor move— R&D recalls it before sprint planning
Support logs a recurring bug— Engineering gets the signal, no ticket needed
Legal flags a compliance issue— every fleet sees it, scoped by permissions
Architect makes a design decision— builders recall it weeks later automatically
Built for the hyper-agent generation
Fleets of agents that learn, adapt, and improve themselves over time
Self-improving agents
Every task makes the next one smarter. Agents write back what they learn—memory compounds, performance climbs.
Cross-fleet intelligence
A discovery in one fleet becomes institutional knowledge for all. No point-to-point wiring needed.
Autonomous curation
Contradictions detected, duplicates merged, noise crystallized into verified knowledge—continuously and automatically.
Governed autonomy
Full read/write freedom within trust levels. Audit trails and fleet boundaries keep autonomy safe at scale.
Why MemClaw is different
Most memory tools are vector databases with a wrapper. MemClaw is a governed knowledge system built for multi-agent teams.
Multi-agent native, not single-agent memory
Memory is shared across fleets—not tied to one session or one agent. Agents learn from each other by default.
Zero-effort writes, full LLM enrichment
Agents send raw text. MemClaw auto-classifies type, extracts entities, scores importance, detects PII, and embeds—all in one call.
Governed by default
Tenant isolation, fleet boundaries, 4-tier agent trust, visibility scopes, and full audit trails on every operation. Not bolted on—built in.
Self-healing contradictions
Conflicting memories detected via RDF triples and LLM analysis. Old facts superseded, duplicates blocked, knowledge stays clean.
Vector + graph + structure in one query
Semantic search, entity knowledge graphs, and structured RDF triples fused into a single retrieval path. Entities auto-extracted on every write.
Agents fine-tune their own recall
Each agent optimizes its search profile—top_k, similarity threshold, keyword blend, graph depth. Retrieval improves per agent over time.
Memory has a lifecycle
8 statuses track every memory: active, confirmed, conflicted, archived, expired. Lifecycle automation cleans up stale and outdated knowledge.
Memory crystallizer
LLM-powered consolidation merges near-duplicate clusters into clean atomic facts. Source memories archived with full provenance.
MCP-native integration
Built-in MCP server and OpenClaw plugin. Claude Desktop, Claude Code, Cursor—connect with a URL and API key. No install.
Compounds over time
Memories persist, update, and accumulate. Recall boost rewards frequently-used knowledge. Your fleet gets smarter the longer it runs.
Watch two fleets share knowledge in real time
The demo sandbox has 6 fleets, 27 agents, and hundreds of memories. Search across fleets, explore the knowledge graph, and see contradiction detection in action.
Architecture
MemClaw combines a vector store, knowledge graph, and LLM enrichment pipeline into a single governed platform. Every write is auto-enriched—classified, entity-extracted, contradiction-checked, and embedded—before landing in the shared memory space. Search blends semantic similarity with graph traversal across fleet boundaries, all governed by tenant isolation, agent trust levels, and full audit trails.
View full architecture diagram →Features
Everything your agents need to build long-term, shared memory
Governance & Sharing
Governed Knowledge Sharing
Visibility scopes (agent, team, org) + 4-tier agent trust levels control who sees what. Cross-fleet sharing is trust-gated and audit-logged. Agents auto-register on first write.
Contradiction Detection & Dedup
Conflicting memories detected via RDF triples and LLM analysis. Duplicates blocked by content hash. Old facts superseded automatically—no silent drift.
LLM Enrichment & PII Detection
Every write auto-classified: 12 memory types, importance score, title, summary, tags, temporal dates. Long content auto-chunked. PII flagged for compliance.
Knowledge & Intelligence
Knowledge Graph & Entity Extraction
People, orgs, and technologies auto-extracted into a live graph. Search expands through relations up to 2 hops. Fuzzy resolution merges duplicates like “OpenAI” and “Open AI”.
Hybrid Search & Recall Briefings
Vector + keyword search with composite ranking: similarity, weight, freshness, graph boost. One-call recall briefings return LLM-synthesized context paragraphs.
Per-Agent Search Tuning
Each agent optimizes its own retrieval: top_k, similarity threshold, keyword blend, graph depth, freshness decay. Retrieval quality improves per agent over time.
Memory Quality
Memory Lifecycle
8 statuses track every memory: active, confirmed, conflicted, archived, expired. Lifecycle automation cleans up stale knowledge. Edits trigger re-embedding with full audit diffs.
Memory Crystallizer
LLM-powered consolidation merges near-duplicate clusters into clean atomic facts. On-demand or scheduled. Source memories archived with full provenance.
Document & URL Ingestion
Paste a URL or text, pick a fleet and focus. LLM extracts atomic facts for preview. Selectively commit as tagged memories with source provenance.
Integration & Operations
MCP & OpenClaw Integration
Built-in MCP server for Claude Desktop, Claude Code, Cursor. OpenClaw plugin with one-liner install, fleet UI, OTA deploys, and agent education.
REST API & Observability
Full CRUD, search, entities, graph. Every operation audit-logged. Latency tracking across MCP, REST, and plugin. OpenAPI docs at /api/docs.
Multi-Tenant & Configurable
Full tenant isolation with per-tenant LLM provider overrides. Toggle graph retrieval, recall boost, and auto-crystallize independently per organization.
Connect in 30 seconds
Two integration paths—MCP for any AI client, OpenClaw plugin for fleet deployments
{
"mcpServers": {
"memclaw": {
"url": "https://memclaw.net/mcp/",
"headers": {
"X-API-Key": "mc_your_key"
}
}
}
}# SSH into your gateway, then: curl -s "https://memclaw.net/api/install-plugin\ ?api_key=mc_key&fleet_id=fleet-001" | bash # Restart OpenClaw: openclaw gateway restart # Installs plugin, builds, configures # allowlist, and sets up heartbeat. # Or use Fleet UI for point-and-click.
memclaw_write MCP + Plugin Send content — LLM auto-enriches memclaw_search MCP + Plugin Semantic + keyword hybrid search memclaw_brief MCP + Plugin LLM-synthesized context briefing memclaw_update MCP + Plugin Update content, type, weight, status memclaw_transition MCP + Plugin active → confirmed → outdated memclaw_entity_get MCP + Plugin Entity with relations & memories memclaw_tune MCP + Plugin Per-agent search retrieval tuning memclaw_delete MCP only Soft-delete by memory ID
Pricing
Start free. Scale as your fleet remembers more.
About MemClaw
MemClaw is built by Caura.ai as the governed memory platform for multi-agent AI systems. From semantic retrieval to knowledge graphs to cross-fleet sharing—MemClaw lets your agents remember, learn, and collaborate across teams with enterprise-grade permissions and audit trails.