REDDIT · REDDIT// 27d agoOPENSOURCE RELEASE

Memento adds durable repo memory for AI agents

Memento is a local-first MCP server written in Go that gives AI coding agents persistent, structured memory about a software repository — storing architecture notes, module relationships, and design decisions that survive across sessions. It exposes MCP tools for intent-routed code retrieval and explicit note storage, with no embeddings, no cloud, and no API keys required.

// ANALYSIS

The context-window amnesia problem is real and annoying, and Memento takes a refreshingly pragmatic approach — explicit notes and deterministic indexing instead of the vector-search-over-everything trap.

–Language-aware import graphs (Go, TS/JS, PHP) let agents navigate large repos without brute-force full-file injection
–Intent-routed retrieval (`navigate`, `implement`, `review`) means the context served matches what the agent is actually trying to do
–No embeddings by design: keeps the privacy story clean, eliminates external API dependencies, and avoids the staleness problem of semantic indices
–VS Code extension auto-downloads the platform binary and handles MCP config setup — low-friction onboarding for a Go CLI tool
–Very early stage (v0.5.0, ~2 months old, 1 GitHub star) and self-described as a scaffold; expect rough edges, but the architecture is coherent

// TAGS

mementomcpopen-sourcedevtoolagentllmide

DISCOVERED

27d ago

2026-03-15

PUBLISHED

28d ago

2026-03-15

RELEVANCE

7/ 10

AUTHOR

caiowilson