Blog
How memory-based AI systems work — retrieval, consolidation, decay, and the architecture decisions behind giving agents memory that lasts.
A Backroom That Remembers: Agent Memory Across a 38-Week Season
Three agents pick one team, every week, for thirty-eight weeks. What they say to each other, what gets written down, and what the next gameweek is able to recall.
What Is AI Agent Memory? A Practical Guide
How memory-based AI systems work: the difference between context and memory, the four stages of the memory lifecycle, and what a production memory layer has to do.
Hybrid Search for Agent Memory: Vectors, BM25 and RRF
Why pure vector search fails on identifiers and exact phrases, how BM25 covers the gap, and how Reciprocal Rank Fusion combines the two into one ranking.
Memory Consolidation and Decay: Keeping Agent Memory Useful
An agent memory store that only ever grows becomes unusable. How deduplication, compression, importance decay and retention zones keep it accurate over time.
Local-First Agent Memory: The Case for One File Per Agent
What you get when an agent's entire memory — vector index, keyword index, knowledge graph and audit log — lives in a single portable file on your own disk.
Giving Coding Agents Memory That Survives the Session
Coding agents forget every architectural decision when the context window compacts. What persistent memory changes, and how to wire it into an MCP client.
New posts, in your inbox
Occasional writing on agent memory — retrieval, consolidation, and what we learn running it in production. No marketing, unsubscribe any time.
Prefer a reader? Subscribe by RSS.