Shared local memory for AI coding tools that stays true to your code.
MCP Server: io.github.anishchugh05/aether-cache
This MCP server provides “Shared local memory for AI coding tools” focused on keeping memory aligned with the user’s code. The server description indicates it is intended for AI coding tool workflows, where locally shared state can be reused while remaining consistent with the current codebase.
🛠️ Key Features
Shared local memory for AI coding tools
Memory intended to stay true to your code
Server name: io.github.anishchugh05/aether-cache
🚀 Use Cases
AI coding tool sessions that need shared local memory
Workflows where maintaining code-consistent state matters
⚡ Developer Benefits
Predictable local state behavior tied to the code
Reuse of memory across AI coding tooling interactions within the local environment
⚠️ Limitations
Limited published details beyond the provided description; no additional capabilities, tools, or configuration data were provided
Claude Code, Cursor, and other AI tools each forget your project between sessions, and when they do remember things,
nothing checks whether those memories are still correct. Mnemolith gives every tool one shared, local memory, and keeps
it honest:
Memories go stale when the code changes. Each memory is linked to the files it's about, with a snapshot of their
git content hash. When docker-compose.yml changes, the memory about it is retired automatically instead of being
fed to your AI forever.
Contradictions get replaced, not hoarded. "We migrated to MySQL" retires "we use Postgres" instead of sitting
next to it and confusing the model. Near-identical repeats are merged.
Every memory can explain itself. Which tool and session saved it, which file versions it was based on, and every
change since: created, seen again, went stale, confirmed, replaced.
It works with the files you already use. Current memories sync into a managed section of AGENTS.md and
CLAUDE.md, so they're committed, reviewed in PRs, and visible to teammates who don't run Mnemolith.
It runs entirely on your machine, on top of a database engine I wrote from scratch.
Not a Redis clone
Mnemolith started as a from-scratch rebuild of Redis in Java, then kept going. Every layer here is written from
scratch, and each one does something the layer beneath it can't:
Layer
What it adds
MCP server
Claude Code, Cursor, and any MCP tool share one memory
Git awareness
Memories retire themselves when the code they describe changes
Memory layer
Status, provenance, full history, duplicate merging, and conflict review, inside the engine
HNSW vector index
Semantic search: 99.4% recall at 0.84 ms on 100K vectors, 22× faster than exact search
Redis-compatible engine
RESP protocol, 50+ commands, replication, transactions, and crash-safe persistence (0 acknowledged writes lost across 1,039,890 writes and 8 kill -9 crashes), at 88–93% of real Redis's throughput
The bottom layer alone is a working Redis: redis-cli, redis-benchmark, and redis-py talk to it unchanged. The
memory layer, git awareness, and MCP server are things Redis doesn't do at all.
code
Claude Code ─┐ ┌─────────────────────────────────────────────┐
Cursor ──────┼─ MCP ─► mnemolith ─┼─ RESP ─► Java engine (Main.java) │
other tools ─┘ (Python: embeddings,│ memories, HNSW vector index, │
git hashing, sync) │ history, append-only file on disk │
└─────────────────────────────────────────────┘
Does it help?
We tested it. Claude Code did 7 tasks in a sample codebase, 3 times each, in three setups. Every task needs
something only a teammate would know: a past decision, who owns which code, a promise to a partner, a business
rule. Two need a fact mentioned in an earlier session. An automatic check grades every run.
Setup
Tasks done right
Across sessions
Avg cost per task
No memory
43% (9 of 21)
3 of 6
$0.12
CLAUDE.md notes, same facts
90% (19 of 21)
4 of 6
$0.12
Mnemolith, same facts
100% (21 of 21)
6 of 6
$0.14
Without memory, Claude edited another team's code, changed a frozen partner API and skipped refund approval, every
time. A CLAUDE.md file held the same facts but missed a database address mentioned in an earlier session 2 times
out of 3; Mnemolith saved it on its own and used it next session. The cost is about 15% more per task than a notes
file, at API prices, because recalling and keeping memories up to date takes a few extra turns. 63 runs on Claude Opus 5.5, November 2026.
In an earlier round where every fact could also be found in the code, all three setups tied.
Run it yourself (uses your own Claude Code plan; see eval/ for what each task checks):
The dashboard (mnemolith ui): memories whose files changed wait for a decision, with the file versions side by side.
And from your AI tool's side, a memory linked to a file, the file changes, and the next lookup catches it:
code
> remember("The database runs Postgres via docker compose", files=["docker-compose.yml"])
Saved as m1.
... someone edits docker-compose.yml ...
> recall("database setup")
These memories were just retired because files they depend on changed. If one is still true,
call confirm_memory; if it's wrong, save the corrected fact with remember:
[m1] "The database runs Postgres via docker compose" is now marked outdated: docker-compose.yml changed.
> confirm_memory("m1") # still true, so re-link it to the new version of the file
Confirmed m1; it's active again and linked to the current file versions.
> explain_memory("m1")
m1: "The database runs Postgres via docker compose"
status: active, project: demo
saved by: claude-code (session eea54b5c)
linked files: docker-compose.yml@02ea346...
history:
19:16 created: by claude-code (session eea54b5c), linked to docker-compose.yml@87da641...
19:16 active -> stale: docker-compose.yml changed since this was saved
19:16 re-linked: to docker-compose.yml@02ea346... (confirmed still true)
19:16 stale -> active: confirmed still true
A newer fact replacing an older one:
code
> remember("The API timeout was raised to 60 seconds")
Saved as m3.
Related existing memories:
[m1] (similarity 0.58) The API timeout is 30 seconds
If m3 makes any of these no longer true, call replace_memory(old_id, "m3"). Otherwise do nothing.
> replace_memory("m1", "m3")
m1 retired; m3 replaces it.
Quick start
One line installs everything and runs setup. On macOS or Linux:
The installer (install.sh, install.ps1) gets uv if you
don't have it (uv brings Python), installs Mnemolith, and runs mnemolith setup. Prefer your own tools? Any of
these works, followed by mnemolith setup:
setup starts the engine in the background, downloads the embedding model once (about 90 MB), and connects every
supported AI tool it finds on your machine: Claude Code, Claude Desktop, Cursor, and Codex. That's it. Open your
AI tool in any git repo and try:
Remember that our tests need DOCKER_HOST set. It depends on docker-compose.yml.
Then start a new session and ask how to run the tests.
Only use Claude Code? You can install Mnemolith as a plugin instead, with no setup step (needs
uv):
The plugin (plugin/) runs the same server and adds a /mnemolith:dashboard command. Use the plugin or
setup, not both, or Claude sees the tools twice.
The engine runs on Java 11+. If you have it, it's used. If you don't, Mnemolith downloads a private copy (about
50 MB) into ~/.mnemolith/java, checks it against its published checksum, and uses it only for Mnemolith;
nothing is installed system-wide, and mnemolith uninstall removes it.
Already keep notes for your AI in CLAUDE.md, AGENTS.md, or Cursor rules? Run mnemolith import in the repo
(or use Import notes in the dashboard) to turn them into memories, linked to the files they mention.
Memories are kept per project (the git repo's folder name), so projects never mix. Everything stays on your machine:
the engine only accepts connections from it, and your memories live in ~/.mnemolith.
Command
What it does
mnemolith setup
One-time setup: engine, model, and AI tools
mnemolith add
Adds the repo you're in (or a path) so the dashboard can check its files; the dashboard can also find and add repos for you
mnemolith ui
Opens the dashboard: see what your AI remembers, and confirm, fix or retire memories whose files changed
mnemolith import
Turns the notes in CLAUDE.md, AGENTS.md, and Cursor rules into memories; shows them first and asks
mnemolith remove
Removes a project (default: the one you're in), like Remove project in the dashboard: its memories are retired, not deleted
mnemolith export
Saves memories, with their linked files and history, to a JSON file (-p for one project)
mnemolith status
Is it running, and what's in this project's memory
mnemolith doctor
Checks everything and tells you how to fix what's wrong
mnemolith list / recall "query" / explain m1
Look at memories from the terminal
mnemolith sync
Writes current memories into AGENTS.md / CLAUDE.md
mnemolith mender on / off
Turns Mender on or off; mnemolith mend drafts fixes for this repo now
mnemolith connect cursor
Connects one tool: claude-code, claude-desktop, cursor or codex (--print shows config for any other MCP tool)
mnemolith start / stop
Runs or stops the engine (it also starts on its own when a tool needs it)
mnemolith uninstall
Disconnects your AI tools, stops the engine, and asks before deleting memories
If mnemolith isn't found after installing, run uv tool update-shell and open a new terminal. Port 7380 taken?
Set MNEMOLITH_PORT and run setup again. To uninstall: mnemolith uninstall, then uv tool uninstall mnemolith.
Mender: drafted fixes for stale memories
Flagging a stale memory is half the job: someone still has to work out what's true now and type it in. Mender, an
optional part of Mnemolith, drafts that for you. mnemolith setup asks whether to turn it on; mnemolith mender on
(or Settings in the dashboard) changes it later. It needs no API key and installs nothing extra.
Simple changes are drafted by plain rules, on your machine. Mender compares the version of each file the memory
was saved against (from git) with the file now. image: postgres:16 became image: mysql:8, so "The database is
Postgres (postgres:16 image)" becomes "The database is MySQL (mysql:8 image)". It handles version bumps, moved files
(the path is updated, not treated as a deletion), and changes that don't touch the memory at all (only a comment, or
lines it doesn't mention), which it suggests confirming as still true.
Everything else is drafted by the AI tool you already use. With Mender on, Claude Code, Cursor, and Codex get two
more tools: list_stale shows each stale memory with what changed in its files, and propose_fix submits a
draft. Drafts are checked against the code: quotes in the reason and new numbers or versions must actually be in
the files, or the draft is marked Unverified.
You approve every change in the dashboard, where each draft sits under the diff it came from, with what it's
based on. Approve it, edit it, or dismiss it. Approving is refused if the files changed again after the draft was
made (it would be based on old code), and if the memory was already settled some other way.
Changed your mind? Approving shows an Undo, and any fix can be reverted later from the memory's details in the
dashboard (Revert this fix) or with mnemolith mender revert m7: the old memory comes back for you to decide,
and Mender won't suggest the same fix again unless the files change. mnemolith status and your AI tool tell you
when fixes are waiting.
mnemolith mend drafts fixes for the repo you're in and lists them.
Moving from Aether Cache
Mnemolith used to be called Aether Cache. If you installed aether-cache, run the installer above. It installs
Mnemolith, removes aether-cache, moves your memories from ~/.aether-cache to ~/.mnemolith, and reconnects your AI
tools. Nothing is lost, and memories synced into CLAUDE.md / AGENTS.md are picked up where they are.
Installed some other way? Install mnemolith the same way, run mnemolith setup, then remove aether-cache
(uv tool uninstall aether-cache, pipx uninstall aether-cache, or pip uninstall aether-cache).
Tools the AI can call
Tool
What it does
remember(fact, files?, replaces?)
Saves a durable fact, optionally linked to files. Merges near-identical repeats and shows related memories so the AI can decide whether the new fact replaces one; replaces="m42" saves it and retires m42 in one call.
recall(query, limit?)
Searches the project's memories by meaning. Checks linked files first, so it never returns a memory whose code changed underneath it.
explain_memory(id)
Where a memory came from and its full history.
replace_memory(old, new)
Retires an older memory that a newer one makes untrue.
mark_outdated(id, reason)
Retires a memory that turned out to be wrong.
confirm_memory(id)
Brings back a memory that's still true after its files changed, re-linked to the current versions.
check_stale()
Runs the file check on demand.
list_conflicts() / resolve_conflict(...)
Reviews pairs of very similar memories that may contradict each other.
sync_to_files()
Writes current memories into AGENTS.md / CLAUDE.md (also: mnemolith sync).
list_stale()
With Mender on: the stale memories, what changed in their files, and which already have a draft.
With Mender on: drafts a fix for the user to approve in the dashboard. Checked against the code before it's shown.
Design decisions, and what the measurements said
Similarity can't detect contradictions, so the AI judges
The first version flagged a new memory as a possible contradiction when it was similar to an existing one. Calibrating
with real all-MiniLM-L6-v2 embeddings showed that can't work:
Pair
Kind
Similarity
"we use Postgres" / "the database is Postgres"
duplicate
0.73
"tests need DOCKER_HOST set" / "you have to set DOCKER_HOST before running tests"
duplicate
0.92
"we use Postgres" / "we migrated to MySQL"
contradiction
0.30
"API timeout is 30 seconds" / "API timeout is 60 seconds"
contradiction
0.94
"tests need DOCKER_HOST set" / "integration tests need a running Redis"
related, both true
0.34
Duplicates scored 0.73 to 0.92 and contradictions 0.30 to 0.94, so no threshold separates them. A threshold low enough
to merge real duplicates would silently merge "timeout is 30s" with "timeout is 60s", the worst bug this product could
have. So the engine only auto-merges near-identical text (similarity ≥ 0.97). For everything else, embeddings find
the related memories and the AI, which already understands the sentences, decides whether one replaces another.
Staleness compares file contents, not commits
Each linked file is stored as path@<git blob hash>, computed with git hash-object. Comparing contents rather than
commit ids catches uncommitted edits, ignores commits that didn't touch the file, and needs no git hooks: recall
re-hashes the handful of linked files before every search.
Memories never mutate silently
Every change is a logged event with a timestamp and a reason. The engine logs each memory with the id and timestamp
it assigned, so replaying the log after a restart rebuilds the exact same memories and histories, down to the
millisecond, instead of minting new ones.
The engine underneath
A Redis-compatible server written from scratch in Java. It speaks RESP, so redis-cli, redis-benchmark, and
redis-py work with it unchanged. The core was built by completing every stage and extension of the
CodeCrafters "Build Your Own Redis" challenge; the vector index,
persistence hardening, and the memory layer are my own additions.
Typical latency is on par with Redis. The gap is in worst-case latency, from JVM garbage collection and JIT warm-up,
which Redis (written in C) doesn't have.
HNSW vector search vs. exact search
100,000 AG News headlines embedded with all-MiniLM-L6-v2 (384 dimensions, cosine similarity), 500 held-out queries,
k = 10. Recall is measured against exact brute-force search on the same data.
Mode
Recall@10
p50 latency
p99 latency
Exact (brute force)
1.000
18.58 ms
20.52 ms
HNSW, ef=10
0.905
0.49 ms
1.38 ms
HNSW, ef=20
0.951
0.52 ms
0.73 ms
HNSW, ef=50
0.985
0.66 ms
0.89 ms
HNSW, ef=100
0.994
0.84 ms
1.41 ms
HNSW, ef=200
0.998
1.24 ms
2.19 ms
At ef=100, HNSW returns 99.4% of the true top-10 about 22× faster than exact search. Building the index for 100K
vectors took 136 seconds (M=16, M0=32, efConstruction=200). Latencies are measured from the Python client, so they
include network and client overhead. Full setup and raw numbers: results.json.
How it works: each vector links to its nearest neighbors in a layered graph. A search starts on a sparse top layer
of long-range links, hops greedily toward the query, drops down a layer at a time, and finishes with a best-first search
on the bottom layer that keeps the ef best candidates. Neighbor selection uses the diversity heuristic from the HNSW
paper (prefer neighbors in different directions over near-duplicates), which is what keeps recall high.
Filtered search: HNSW doesn't know about memory status, so searches over-fetch candidates, drop the ones that don't
match (stale or superseded), widen the search if too few survive, and fall back to an exact scan if needed. In a test
with 2,000 memories where only 2% were active, filtered searches returned exactly the correct top 10 on 30/30 queries.
The catch: at that selectivity, the exact fallback runs most of the time, which is the known weak spot of
post-filtering on graph indexes.
Crash safety
Every write goes to an append-only file (AOF) that is replayed on startup.
kill -9 crash test (crash_test.py): a client writes as fast as it can, the server is killed at a random moment,
restarted, and every write the server acknowledged is checked.
fsync policy
Crashes
Acknowledged writes
Lost
everysec
5
757,295
0
always
3
282,595
0
Recovery took 0.6 seconds per restart, replaying up to ~185K writes.
Torn writes: a machine crash can cut the last write off halfway. The first version refused to start on such a file;
now replay stops at the last complete command, trims the partial one, and starts normally with all earlier data intact.
Cost of each fsync policy (SET throughput, 50 clients):
Policy
Throughput
p50 latency
always (sync every write)
42.9K/sec
1.159 ms
everysec (sync once a second)
113.9K/sec
0.103 ms
no (let the OS decide)
118.8K/sec
0.103 ms
everysec is 2.7× faster than always and within 4% of no, which is why it's the sensible default. On macOS, fsync
doesn't force the SSD's own cache to flush, so always looks cheaper here than it would on a Linux server.
Full command list
Memory:
MEM.ADD, MEM.GET, MEM.SEARCH (with STATUS filter), MEM.STATUS, MEM.WHY, MEM.LIST
java src/main/java/Main.java # port 6379
java src/main/java/Main.java --bind 127.0.0.1 # accept connections from this machine only
java src/main/java/Main.java --port 6380 --replicaof "localhost 6379"
java src/main/java/Main.java --dir /tmp/data --appendonly yes --appendfsync everysec
Reproducing the benchmarks
Each benchmark runs against the engine you've started:
sh
redis-benchmark -p 6379 -t set,get # throughput; also run against redis-server to compare
python vector_bench.py # HNSW recall and latency (start the engine with -Xmx4g)
python crash_test.py everysec 5 # kill -9 crash test
python memory_mcp/calibrate.py # embedding similarity calibration
Related work
Mnemolith is an early, open-source project. These are the production systems in the same space:
Instruction files:AGENTS.md and CLAUDE.md. Mnemolith syncs into these rather than
replacing them.
Roadmap
What's coming next, roughly in order. Ideas and pushback are welcome in
GitHub issues.
Next: team memory (December to February)
Shared memory for a repo. Each person keeps their own memories, and each repo gets one shared team memory that
every teammate's AI reads. You promote a memory to the team explicitly, so nothing gets shared by accident.
Reviewed like code. Team memories live in the repo as plain files. Mnemolith writes the file; you commit it
and open the pull request yourself, so teammates review a change to team knowledge the same way they review code.
Git history is the audit trail, and access follows repo permissions. Mnemolith never commits on its own.
Stale memories flagged in pull requests. A GitHub Action comments when a pull request changes files that team
memories depend on, so outdated knowledge gets caught in review.
Then: memory for any folder (early 2027)
Memories for folders that aren't code, tied to PDFs, Word files and spreadsheets, and flagged when those change,
so people who aren't developers can use it too.
Later, a desktop app that installs like any other app, with no terminal needed.
Always: the local version stays free and open source.
Status and limitations
Mnemolith is v1 and early. It works end to end in automated tests and is now being tried in real Claude Code sessions.
Not done yet:
An evaluation on real-world repos. The evaluation above uses a small sample codebase built for
it; real projects are bigger and messier.
Lower overhead: recalling and keeping memories up to date costs about 15% more per task than a notes file.
AOF compaction: the log only grows, and every status change adds a line.
Engine limitations:
Threading: one OS thread per connection, not an event loop. That's fine for a local tool, but it won't scale to
thousands of clients like Redis.
Transactions:EXEC isn't isolated, so other clients' commands can interleave with a transaction's commands.
Vector index: re-adding a vector updates it without rewiring its graph links, and there's no delete. Retired
memories are filtered out, not removed.
Replication: full resync only, and a new replica assumes the master's dataset is empty.
Smaller gaps:
KEYS * lists only string keys.
GEOSEARCH scans every member.
ACL supports only the default user.
HELLO (RESP3) isn't supported, so clients must use RESP2.
Project layout
code
src/main/java/Main.java the engine (single file, from the CodeCrafters setup)
mnemolith/ the Python package behind the `mnemolith` command
server.py MCP server: embeddings, git staleness, AGENTS.md/CLAUDE.md sync
engine.py finds Java, starts and stops the engine in the background
javart.py downloads a private copy of Java when the computer has none
connect.py registers Mnemolith with Claude Code, Claude Desktop, Cursor, and Codex
cli.py the `mnemolith` command
ui.py, ui/index.html the dashboard (`mnemolith ui`): a local server, its API, and the page
mender.py Mender: rule-drafted fixes, drafts from AI tools, and the guarded approve
portability.py import from notes files, export to JSON
engine/mnemolith-engine.jar the prebuilt engine (rebuild with `python scripts/build_engine.py`)
tests/test_mender.py Mender's rules (`python -m pytest tests`)
eval/ with-vs-without memory evaluation on a sample codebase (`python3 eval/run.py`)
plugin/ the Claude Code plugin (listed through .claude-plugin/marketplace.json)
memory_mcp/calibrate.py similarity calibration for duplicate/contradiction thresholds
vector_bench.py HNSW benchmark (writes results.json)
crash_test.py kill -9 crash-safety test
Releasing
Bump the version in pyproject.toml, mnemolith/__init__.py, and server.json, commit, then:
sh
git tag v0.3.4 && git push --tags
Once the release is on PyPI, point the Claude Code plugin at it: bump the version in
plugin/.claude-plugin/plugin.json and plugin/pyproject.toml, run uv lock in plugin/, then commit and push.
The plugin can't move first, because its lockfile needs the new release to exist.
The release workflow checks the versions match, builds and smoke-tests the package,
publishes it to PyPI and the MCP Registry, and creates the GitHub Release.