xmem — Graph-Based Agent Memory for OpenClaw

xmem gives your OpenClaw agent a persistent, graph-based long-term memory.

Instead of simple vector recall, xmem builds a knowledge graph — memories are connected through entities, relationships, and temporal context. Your agent remembers facts, tracks decisions, discovers connections through multi-hop graph traversal, and consolidates knowledge over time.

Quick Start

# Install the plugin
openclaw plugins install @xmem-space/openclaw-plugin

# Sign up and get your API key at https://xmem.space

# Configure
openclaw config set plugins.entries.memtap.config.apiKey "mt_live_your_key_here"
openclaw config set plugins.entries.memtap.config.serverUrl "https://api.xmem.space"

# Restart gateway
openclaw gateway restart

That's it. Your agent now has memory.

What's New in v8.0 — "AST & Intelligence"

Six new tools exposing xmem server v8 features:

| Tool | Description | |------|-------------| | memtap_code | AST-based code analysis — extract symbols and call graphs from source files | | memtap_transcribe | Transcribe audio/video files (mp3, m4a, wav, mp4, etc.) to text | | memtap_communities | Louvain community detection — discover entity clusters in the graph | | memtap_graph_hygiene | Detect god-nodes (over-connected entities) that pollute the graph | | memtap_surprising | Find surprising connections: semantically close but unlinked entity pairs | | memtap_graph_report | Full graph health report (entity counts, density, communities, etc.) |

Hook enhancements:

• Audio/video files captured by the agent are auto-transcribed and stored as searchable memories
• Code files are analyzed via AST endpoint for richer symbol extraction (regex fallback preserved)

All v7.4 features (SHA-cache, .xmemignore) remain fully backward compatible.

What's New in v7.4 — "Cache & Ignore"

• Persistent SHA256 cache (.xmem-hashes.json at workspace root) — re-scans skip unchanged files, saving ~80–90% of requests when re-onboarding.
• .xmemignore — gitignore-syntax file to exclude paths from the workspace scan (supports ! negation). See scripts/xmemignore.template for a starter.
• memtap_scan { force: true } — bypass the cache and re-send every file.
• Fully backward compatible: without these files, the plugin behaves exactly as v7.3.

What It Does

20 Tools

| Tool | Description | |------|-------------| | memtap_recall | Semantic search across the knowledge graph | | memtap_remember | Store a new memory with auto-extracted entities | | memtap_bulletin | Context bulletin with graph expansion — agent "knows" related things | | memtap_graphrag | Multi-hop GraphRAG: vector/BM25 search + graph traversal | | memtap_maintenance | Memory hygiene: decay reports, contradiction detection, dedup | | memtap_graph | Graph analysis: overview, gaps, clusters, connections, traverse | | memtap_decide | Decision tracking: create, list, resolve, defer | | memtap_memory | CRUD operations on individual memories | | memtap_entities | Entity management: list, get linked memories, merge duplicates | | memtap_edges | Create relationships between memories | | memtap_health | Server health check and memory statistics | | memtap_monitor | Neural performance monitoring and analytics | | memtap_alerts | Anomaly detection and alerting system | | memtap_dashboard | Comprehensive system overview dashboard | | memtap_code | AST code analysis: symbols, call graphs, language stats | | memtap_transcribe | Audio/video transcription to text | | memtap_communities | Louvain community detection in the knowledge graph | | memtap_graph_hygiene | God-node detection for graph health | | memtap_surprising | Discover surprising unlinked connections | | memtap_graph_report | Full graph health report (JSON or Markdown) |

5 Hooks

• Pre-message recall — Automatically loads relevant context before each conversation turn
• Post-message capture — Extracts and stores important information from conversations
• Bootstrap bulletin — Injects a memory context bulletin when an agent session starts
• Periodic consolidation — Background memory maintenance and knowledge consolidation
• Session analytics — Performance monitoring across sessions

Configuration

{
  "plugins": {
    "entries": {
      "memtap": {
        "enabled": true,
        "config": {
          "serverUrl": "https://api.xmem.space",
          "apiKey": "mt_live_...",
          "agentId": "main",              // optional: scope memories per agent
          "autoCapture": true,            // auto-extract memories from conversations
          "bulletinOnBoot": true,         // inject memory context on session start
          "bulletinTopics": ["projects", "preferences"],
          "decayRate": 0.005              // memory importance decay per day
        }
      }
    }
  }
}

Optional: Embedding Support

For vector-enhanced GraphRAG search, configure an embedding provider:

{
  "embeddingUrl": "https://api.openai.com/v1/embeddings",
  "embeddingModel": "text-embedding-3-small",
  "embeddingApiKey": "sk-..."
}

Pricing

| Plan | Price | Memories | Agents | API Calls | |------|-------|----------|--------|-----------| | Free | $0/mo | 1,000 | 1 | 10,000 | | Starter | $13/mo | 50,000 | 3 | 500,000 | | Pro | $43/mo | 500,000 | 15 | 5,000,000 | | Team | $109/mo | 2,000,000 | Unlimited | 50,000,000 | | Enterprise | $1,299/mo | 10,000,000 | Custom | Custom |

Sign up at xmem.space to get your API key.

Why xmem?

• Graph-based: Full knowledge graph, not just vector similarity
• Cloud-native: Managed service — no database setup needed
• GraphRAG: Multi-hop traversal discovers connections vector search misses
• Decision tracking: Unique decision management for AI agents
• Auto-capture: Learns from conversations without explicit commands
• Memory decay: Natural forgetting keeps the graph relevant
• 8x cheaper: $13/mo vs $99/mo (Mem0, Zep)

Links

• Website: xmem.space
• API Docs: api.xmem.space
• GitHub: github.com/psifactory/memtap
• Support: github.com/psifactory/memtap/issues

License

MIT — psifactory LLC