The blog.
Product decisions, release notes, build logs, and short field reports. Less ceremony than research. More durable than a forum thread.
iCog vs mem0 vs Letta (MemGPT): choosing an AI memory layer
Three good projects, three different problems. A factual map of the AI-memory category and where each one actually fits.
mem0, Letta, and iCog all get filed under 'AI memory,' but they answer different questions for different buyers. A fair, non-marketing comparison of the category — agent SDKs vs stateful agent frameworks vs an MCP-native, cross-tool memory layer.
Persistent memory for AI agents: why context windows aren't memory
A long context buys you a longer conversation. It does not buy you an assistant that remembers you tomorrow, or in a different tool.
Context windows are working memory — large, fast, and erased the moment a session ends. Real persistent memory for AI needs durability, recall, decay and consolidation, and cross-tool reach. Here's the difference, and how an MCP memory layer closes the gap.
Community without a forum
The public layer should be curated. The social layer should live inside the product.
iCog needs public writing and product-native collaboration, not a heavy external forum as the center of gravity.
What this blog is for
A public log for product decisions, build notes, and the parts of iCog that should be explained plainly.
This blog is the lightweight public surface for iCog updates: shorter than research, more durable than chat, and easier to own than a traditional forum.