Project Engram is a three-tier memory architecture for desktop AI agents. It replaces flat key-value memory stores with a biologically-inspired system modeled on how human memory works: incoming information flows through a sensory buffer, gets prioritized in working memory, and consolidates into long-term storage with automatic clustering, contradiction detection, and strength decay. The result is agents that remember context across sessions, learn from patterns, and forget gracefully.
This document describes the architecture as implemented in OpenPawz — a Tauri v2 desktop AI platform. The system implements three memory tiers, a persistent graph, hybrid search with reciprocal rank fusion, background consolidation, field-level encryption, and full lifecycle integration across chat, tasks, orchestration, and multi-channel bridges.
This is our passion and looking for more minds to get involved!
This document describes the architecture as implemented in OpenPawz — a Tauri v2 desktop AI platform. The system implements three memory tiers, a persistent graph, hybrid search with reciprocal rank fusion, background consolidation, field-level encryption, and full lifecycle integration across chat, tasks, orchestration, and multi-channel bridges. This is our passion and looking for more minds to get involved!
All code is open source under the MIT License.