Aide-memory – persistent memory for AI coding agents and teams
Path-scoped memory beats project-wide rules files for team context sharing.
Three-path memory for LLM agents: KV window (exact attention) + retrieval index (archived chunks) + TRN recurrent state (8-96 KB, O(1) per step)
Keeps agent memory at 8 KB constant size while KV caches bloat to 156 MB.
LLM developers, Agent builders
RWKV · RetNet · MemGPT
Path-scoped memory beats project-wide rules files for team context sharing.
Primitive heaps beat FibonacciHeap by killing GC overhead in Java.
Injects raw KV tensors directly into model cache to skip 90% of token recomputation.
Git-verified memory ledger beats vague agent context files.
File-based state management beats vector databases for most local AI coding workflows.
Git repo as versioned agent memory — no database, just markdown files.