Skip to content

docs: Agent memory systems research and starred repos analysis#292

Draft
nsheaps wants to merge 17 commits intomainfrom
claude/review-repos-clawhub-7ZNiq
Draft

docs: Agent memory systems research and starred repos analysis#292
nsheaps wants to merge 17 commits intomainfrom
claude/review-repos-clawhub-7ZNiq

Conversation

@nsheaps
Copy link
Copy Markdown
Owner

@nsheaps nsheaps commented Mar 23, 2026

Summary

  • Comprehensive research on agent memory systems, analyzing ~270 starred repos on nsheaps' GitHub
  • Deep dive into ClawHub/OpenClaw ecosystem (13,700+ skills), especially self-improvement and memory skills
  • Comparison of major memory frameworks: Mem0, Letta/MemGPT, Codebase Memory MCP, Claude-Mem, Cognee, MemOS, Zep, LangMem
  • Academic research survey: CoALA taxonomy, Reflexion, Voyager, EvolveR, SAGE, ACE, A-MEM, metacognitive learning
  • Architectural patterns identified: tiered memory (HOT/WARM/COLD), WAL protocol, 3x confirmation rule, progressive disclosure
  • Session retrospective with process improvement recommendations

Research Documents

  • docs/research/agent-memory-systems/README.md — Main report: task memory, cross-session memory, self-improvement, 8 architectural patterns, recommendations
  • docs/research/agent-memory-systems/state-of-the-art-survey.md — Academic research survey with CoALA taxonomy, additional frameworks, self-improvement research, emerging trends
  • docs/research/agent-memory-systems/starred-repos-analysis.md — All ~270 starred repos categorized by theme
  • docs/research/agent-memory-systems/clawhub-openclaw-ecosystem.md — OpenClaw skills deep dive (self-improving agents, proactive agent, claude-mem)
  • docs/research/agent-memory-systems/memory-frameworks-comparison.md — Technical framework comparison table
  • docs/research/agent-memory-systems/session-retrospective.md — What went well/didn't, process improvements

Test plan

  • Review research accuracy and completeness
  • Verify all external links are valid
  • Check for any missing repos or frameworks

https://claude.ai/code/session_0151jgfinR5ApykyrdySBQRE

claude and others added 13 commits March 23, 2026 07:35
Research covering starred repos analysis (~270 repos), ClawHub/OpenClaw
ecosystem deep dive, and comparison of major memory frameworks (Mem0,
Letta/MemGPT, Codebase Memory MCP, Claude-Mem). Includes patterns for
tiered memory, self-improvement, cross-session persistence, and
architectural recommendations for a potential memory plugin.

https://claude.ai/code/session_0151jgfinR5ApykyrdySBQRE
Adds comprehensive survey covering CoALA taxonomy, additional frameworks
(Cognee, MemOS, Zep, LangMem), self-improvement research (Reflexion,
Voyager, EvolveR, SAGE, ACE, A-MEM), MCP memory servers, and 10
emerging trends in agent memory architecture.

https://claude.ai/code/session_0151jgfinR5ApykyrdySBQRE
Verified 257 repos across 37 categories. Added distribution table,
top starred repos by relevance, and notable repos not in initial
analysis (OpenViking, opencrabs, yoyo-evolve, lemonai).

https://claude.ai/code/session_0151jgfinR5ApykyrdySBQRE
Adds three major findings: Cortex Memory (3-tier cognitive memory with
Ebbinghaus forgetting curve, 93% Recall@1), BrainX (multi-agent shared
memory with 15-step daily pipeline, contradiction detection), and the
ClawHavoc security incident (1,184 malicious skills targeting memory
files). Also documents OpenClaw's native memory system.

https://claude.ai/code/session_0151jgfinR5ApykyrdySBQRE
Adds BM25+vector scoring formula (70/30 weights, union not intersection,
MMR diversity re-ranking), LanceDB memory plugin with enhanced
memory-lancedb-pro variant (RRF fusion + cross-encoder rerank).

https://claude.ai/code/session_0151jgfinR5ApykyrdySBQRE
Adds Hindsight (MCP-native memory, 91.4% LongMemEval, 4 memory networks,
mental models) and 5 key architectural insights including prompt
optimization as underexplored frontier and MCP as integration standard.

https://claude.ai/code/session_0151jgfinR5ApykyrdySBQRE
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants