NetClaw Gets Long-Term Memory: Community Contribution Brings MemPalace Integration
One of the most exciting aspects of open source is when the community steps up and adds capabilities you didn’t even know you needed. Today we’re thrilled to announce that NetClaw now has long-term memory — thanks to a fantastic contribution from community member Satyam Thakur.
The Memory Problem
AI agents are brilliant in the moment, but they forget everything between sessions. Ask NetClaw to troubleshoot a network issue today, and tomorrow it has no memory of what you discussed, what you tried, or what worked.
OpenClaw already had short-term memory through daily log files (memory/YYYY-MM-DD.md), but these are ephemeral — yesterday’s context fades as today’s work takes over.
Enter MemPalace: A Two-Layer Memory Architecture
The new integration creates a two-layer memory system:
| Layer | Purpose | Technology |
|---|---|---|
| Short-Term | Daily session logs | OpenClaw’s existing memory/YYYY-MM-DD.md files |
| Long-Term | Persistent knowledge across sessions | MemPalace MCP with temporal knowledge graphs |
What is a Memory Palace?
MemPalace uses a spatial metaphor inspired by ancient rhetoric techniques. Instead of a flat search index, memories are organized into a navigable structure:
- Wings — People or projects (e.g., "datacenter-migration", "customer-acme")
- Rooms — Specific topics within a wing
- Halls — Memory types (facts, events, relationships)
- Closets — Summaries for quick retrieval
- Drawers — Original verbatim content
This architecture means NetClaw can now remember that "the BGP peer with AS65000 was flapping last month due to MTU issues" and recall that context when similar symptoms appear again.
19 Tools for Memory Management
The MemPalace MCP server provides 19 tools across five categories:
Palace Operations
- Check memory status
- List wings and rooms
- Semantic search with filtering
- Duplicate detection
Temporal Knowledge Graphs
- Store facts with validity windows ("this was true from January to March")
- Query what was true at a specific point in time
- Track entity relationships over time
- Generate timelines
Navigation
- Traverse the knowledge graph across wings
- Discover tunnels between related topics
- Analyze connectivity patterns
Content Management
- Add and organize memories
- Delete outdated information
Agent Diaries
- Per-agent continuous diaries using AAAK compression
- Enables specialized expertise development over time
Why This Matters for Network Operations
Imagine these scenarios:
Recurring Issues: "This interface keeps flapping. Have we seen this before?"
NetClaw can now search its long-term memory and find that yes, this same interface had issues 6 months ago, and the root cause was a bad SFP.
Tribal Knowledge: "What’s the story with VLAN 100?"
NetClaw remembers that VLAN 100 was originally created for guest WiFi, was repurposed for IoT devices in Q2, and has special QoS policies because of that legacy camera system nobody wants to touch.
Change Context: "Why did we set the OSPF cost to 1000 on this link?"
NetClaw recalls the change request from 8 months ago where the link was deprioritized during a circuit migration that never got cleaned up.
Semantic Search Without API Calls
MemPalace uses ChromaDB for semantic search — meaning NetClaw can find relevant memories even when the exact keywords don’t match. Searching for "BGP problems" will find memories about "peering session failures" or "route convergence issues."
And it’s all local. No API calls to external services. Your network’s institutional memory stays on your infrastructure.
Community-Driven Innovation
This integration exists because Satyam Thakur saw an opportunity and contributed PR #55. The changes include:
- MCP server registration in
config/openclaw.json - Installation automation in
scripts/install.sh - A comprehensive skill definition at
workspace/skills/mempalace/SKILL.md - Documentation updates across README, TOOLS, SOUL, and AGENTS files
This is what open source is all about — the community identifying gaps and filling them with elegant solutions.
Getting Started
After updating NetClaw, the MemPalace MCP server is available immediately. NetClaw will automatically learn the memory protocol and AAAK dialect from status responses — no manual configuration needed.
Start building institutional memory by simply working with NetClaw. Over time, your AI network assistant will develop deep contextual knowledge about your specific environment, your historical decisions, and your operational patterns.
Thank you to Satyam Thakur for this contribution. NetClaw continues to grow through community involvement — now with long-term memory capabilities that transform ephemeral AI assistance into persistent institutional knowledge.