Field notes on machine memory
Building iCog in public: persistent memory, MCP, and the messy engineering behind an AI that actually remembers you. For the deep, peer-reviewed-grade analysis, see iCog Research.
AI Memory That Learns User Patterns: The Mem0 Gap
Mem0 stores and retrieves facts. It doesn't learn implicit patterns from repeated user behavior across sessions. Here's the gap, and how CognitiveX closes it.
Memory, explainedAn AI Second Brain That Learns, Not Just Stores
Most second-brain apps store the notes you write. iCog learns who you are across your apps, finds patterns about your life, and shows its work.
Memory, explainedI Want an AI That Remembers Me
Tired of re-explaining yourself every session? Here is how an AI that builds a persistent, editable model of who you are works, and what you should own.
ComparisonsBest Mem0 Alternatives in 2026
An honest ranking of the best Mem0 alternatives in 2026, CognitiveX, Zep, Letta, Cognee, Supermemory, and the learning axis every one of them omits.
Memory, explainedA ChatGPT Memory Alternative You Own
ChatGPT memory is real but locked to one app, capped at ~1,400 words, and not portable. Here's a memory you own across every AI tool, via MCP.
IntegrationsCrewAI External Memory: Wire CognitiveX as the Backend
Outgrew CrewAI's native memory? Use the ExternalMemory provider interface to wire CognitiveX as a hosted, salience-ranked memory backend. Code + honest limits.
Memory, explainedDo LLMs Have Memory?
No, LLMs are stateless and forget everything between calls. Here's why, how ChatGPT and Claude 'remember' anyway, and what memory really is.
Memory, explainedEpisodic vs Semantic Memory in AI
Episodic memory stores events; semantic stores facts. Here's the difference, how it maps to AI memory systems, and why the promotion between them matters.
BenchmarksLongMemEval: The AI Memory Benchmark, Explained
LongMemEval tests long-term, multi-session AI memory across 500 questions. Here's what it measures, the scores models get, and how to read claims skeptically.
IntegrationsMCP Memory Server: Hosted vs Local
Compare the official local MCP memory server with hosted, consolidating, cross-client options. How JSONL knowledge graphs differ from a memory that learns.
ComparisonsThe Mem0 Alternative That Consolidates
Mem0 stores the facts your users state. CognitiveX also learns the patterns they repeat. An honest switcher's guide: when mem0 is fine, and when to move.
ComparisonsCognitiveX vs Supermemory: Storage vs Consolidation
mem0 vs Supermemory, honestly, both store and merge memory; CognitiveX consolidates it. A sourced comparison of connectors, self-hosting, and the dream-pass wedge.
ComparisonsMem0 vs Zep vs Letta vs Cognee vs CognitiveX (2026)
The honest 2026 comparison of AI memory frameworks, mem0, Zep, Letta, Cognee, and CognitiveX, by architecture, self-hosting, temporal recall, and learning.
Memory, explainedMemory Consolidation for AI Agents, Explained
Memory consolidation is how an AI agent restructures what it has stored over time, reinforcement, decay, episodic-to-semantic promotion. Here's how it works.
IntegrationsGive Claude Code Real, Persistent Memory
Claude Code memory caps at 200 lines and stays machine-local. CognitiveX gives Claude Code hosted, cross-session memory that consolidates, installed as an MCP server.
IntegrationsPersistent Memory for Cursor (MCP Plugin)
Cursor forgets your project between chats and removed Memories in v2.1. Add persistent memory for Cursor with CognitiveX as an MCP server that learns.
IntegrationsLong-Term Memory for LangGraph Agents
LangGraph's BaseStore persists JSON by namespace/key but never consolidates. Here's how it works, and how CognitiveX adds a consolidating drop-in memory layer.
Memory, explainedMemory That Learns, Not Just Stores
AI memory that learns goes beyond storage, it consolidates, promoting episodes to facts and decaying noise. Here's why storing isn't remembering.
Memory, explainedPersonal AI Memory: An AI That Learns You
Personal AI memory used to mean storing what you said. The frontier is an AI that learns your patterns across apps, shows its work, and stays yours.
Memory, explainedRAG vs Memory: Why Stateful Isn't Enough
RAG vs memory for LLM agents: RAG is stateless retrieval over a corpus, memory is stateful and per-user. The real question is whether memory consolidates.
Memory, explainedSelf-Improving Agent Memory: Beyond Storage
Self-improving agent memory consolidates instead of just storing. How dream/sleep consolidation works for any model, user-owned, vs Claude Dreaming and Letta.
Memory, explainedThe 4 Types of AI Memory (With a Table)
Working, episodic, semantic, and procedural, the four types of AI memory explained, with a table mapping what each stores, examples, and lifespan.
Memory, explainedWhy ChatGPT Forgets What You Told It
ChatGPT forgets because the model is stateless and its context window is a finite buffer. Here's what's really happening, and how a memory layer fixes it.
ComparisonsThe Zep Alternative With Self-Hosting
Zep killed its self-hostable Community Edition. CognitiveX gives you temporal recall on Postgres + pgvector, no Neo4j, no Graphiti service to operate.
Building in publicWhy we killed our sufficiency check
It looked smart on the whiteboard and added 300ms to every single reply, while timing out 100% of the time. A war-story about latency, reasoning models, and the cost of clever.
Memory, explainedWhat is an AI memory layer?
LLMs are stateless. They forget you the moment a chat ends. An AI memory layer is the system that gives them persistent, long-term memory of who you are. Here's how it works and why it's the missing piece.
ComparisonsiCog vs Mem0 vs Letta vs Zep: choosing an AI memory layer
Four ways to give AI long-term memory, and where each one fits. An honest comparison of the personal, MCP-native, and enterprise approaches to agent memory in 2026.