Echo Research
Notes on memory, context, and agentic work.
Short technical reports from building a memory layer that works across tools instead of making every model re-read the whole past.Context Window Analysis · Jun 26, 2026
How dirty is your context window?
A field report on why agentic coding sessions spend nearly all their tokens re-reading old context, and what that does to cost, speed, and reliability.AI Memory · Jun 2026
Evidence-First Memory: EchoMem Reaches 95.8% on LongMemEval
A benchmark report on reliable AI memory, evidence-backed recall, and what it takes for memory systems to stay accurate across time.Context Engineering · Jun 25, 2026