Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@nyxilum-memory-mcp remember we chose JWT over sessions because of mobile clients"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Detail memory for long AI-coding sessions. An MCP server written in pure
NyxilumLang (.nx) that stores the things a new session
loses: why a choice was made, what was tried and rejected, side findings, non-obvious nuances — and
pulls back only what the current task needs.
Store | Holds |
Architecture docs | WHAT exists (state, stack, structure) |
Claude Code | who the user is, working rules, short facts |
this | WHY: decisions + reasons, rejected options, findings, nuances |
One record = one Markdown file, records/<project>/<id>.md, never rewritten.
State (archived, recall counters) = replay of an append-only events.jsonl.
recall(task, project) — own scoring (tokenize → prefix "stem" → IDF → weights tag×3 / title×2 / body×1,
project bonus, hit boost, noise threshold). Returns a short list within a character budget;
get(ids) returns full text only for what you need.
Cleaning: review_old reports candidates with reasons (not recalled for N days, repo changed a lot
since, linked file vanished). archive is reversible and needs user_confirmed. purge (physical
delete, archived only) is two-step: purge_request returns a one-time phrase the human must type
exactly; purge_confirm compares it character by character (a wrong try = refusal + new phrase).
Hooks: SessionStart injects a tiny digest (counts + recent titles for the current project);
Stop nudges (non-blocking) when many turns passed without remember.
Tools: remember recall get list handoff review_old archive unarchive purge_request purge_confirm.
Related MCP server: arcane
Needs the nx binary (NyxilumNode releases) on disk.
git clone https://github.com/Faneraiy14/nyxilum-memory-mcp
claude mcp add nyxilum-memory -s user -- /path/to/nx /path/to/nyxilum-memory-mcp/server.nxHooks (optional) — add to ~/.claude/settings.json, exec form (no shell):
{"hooks":{
"SessionStart":[{"hooks":[{"type":"command","command":"/path/to/nx","args":["/path/to/nyxilum-memory-mcp/hooks/hook.nx","SessionStart"],"timeout":15}]}],
"Stop":[{"hooks":[{"type":"command","command":"/path/to/nx","args":["/path/to/nyxilum-memory-mcp/hooks/hook.nx","Stop"],"timeout":15}]}]}}Hooks stay silent until the data folder exists (opt-in).
Env | Default | Meaning |
|
| where data lives |
|
| where your repos are (project detection, "repo changed" check) |
<data dir>/config.json overrides any key: stemLength, stopWords, weightTag/Title/Body,
projectBonus, hitBoost, minScore, defaultLimit, defaultMaxChars, staleDaysNoRecall,
staleDaysFinding, staleCommits, purgePhraseMinutes, dupScore, reminderTurns,
reminderQuietMinutes, digestItems, kinds (e.g. add your own record kinds). Defaults: lib/config.nx.
npm install && npm test — unit tests in .nx (tokenizer, IDF, scoring) + end-to-end through the real
MCP client (full lifecycle, purge phrase, broken files, hooks, path traversal, 400-record scale).
MIT.
This server cannot be deployed
Shared memory for coding agents. Stop re-explaining your codebase every session.
Durable, revision-pinned memory storage and retrieval for AI conversations.
Persistent memory for AI agents. Search, store, and recall across sessions.
Shared project memory that keeps teammates and AI agents aligned across sessions.
