Good memory knows when to let go.
Code changes. Decisions change. The context your AI carries into the next session should change with them.
Remembering is only
half the problem.
An AI coding tool starts a new session without the decisions that shaped your project. Saving those decisions helps—until the code moves on and the saved context doesn’t.
Aether Cache connects a fact to the files behind it. When those files change, the fact can leave active memory without disappearing from history. You can see what was believed, what changed, and why.
The goal is context you can inspect, question, and keep current.
Not a Redis clone.
It started as a from-scratch rebuild of Redis in Java. Then it kept going: vector search, traceable memories, and an awareness of the code underneath them.
Every layer written from scratch.
The Redis protocol stays
compatible: redis-cli and redis-py work
unchanged.
-
MCP server
The connection to Claude Code, Cursor, and Codex.
-
Git awareness
Content hashes link memories to files. Changed files retire old facts.
-
Memory layer
Status, provenance, full history, duplicate merging, and conflict review.
-
HNSW index
Its own vector index. 99.4% recall@10 at 0.84 ms on 100K vectors.
-
Java engine
Redis-compatible. 50+ commands, replication, transactions, and crash-safe persistence.
From the tool connection down to the bytes on disk.
A working v0.1.7. Still early.
This is v0.1.7. The engine and memory workflow are here. There’s more to measure, and more to build.
Local shared memory, file-linked retirement, provenance, file
sync, MCP connections, a browser dashboard for reviewing
memories (aether-cache ui), importing the notes
you already keep in CLAUDE.md, AGENTS.md, and Cursor rules,
JSON export, a clean uninstall, and a private copy of Java downloaded
for you when your computer doesn’t have one.
Team memory with GitHub sign-in, so a team shares one project
memory and pull requests that would make memories stale get
flagged.
A with-vs-without-memory evaluation.
Log
compaction.
Built in the open.
A project by Anish Chugh. Read the code, follow the work, or try it locally.