Jev's agent memory works like a brain: dream journals, engrams and spreading activation
RileyRalmuto · x · 2026-11-13
The author describes the original memory system behind the agent Jev, framed as "memory with a metabolism":
- It's a graph-based system built around memory traces called "engrams".
- Overnight "dreaming": the agent gets personal time to journal freely, then consolidates memories and reviews the day's sessions for anything missed or needing correction.
- It draws connections between semantically similar memories automatically.
- Recall uses spreading activation instead of search: initial relevant memories fire when a message arrives, each firing memory triggers connected ones, expanded only two steps out. Users can plug in their own API for this part.
Related event: Agent Jev Introduces Graph-Based 'Memory with a Metabolism'(2 posts)→
More from coding & agent
- Agent marketplace POASwap proposes maker bonds to fix its zero-trade problem — adamamcbride · 2026-11-13
- Sergey Karayev shows Superconductor UI: one ticket, many models in parallel — sergeykarayev · 2026-11-13
- Multiple labs independently discover 'idea basins' where agents get stuck in solution space — AlexGDimakis · 2026-11-13
- SparseRead cuts agent token use by up to 92.9% by deciding what to read before reading — aigclink · 2026-11-13
- Indie dev builds AI design tool Rivermark with Claude in a month, already monetizing via side gigs — JannVanDam · 2026-11-13
- Coding agents on work laptops called a massive security risk after reported Claude Code extension compromise — alvelda · 2026-11-13