The memory layer for Claude Code

Persistent memory and a spine for Claude Code.

Not boilerplate you copy and outgrow — a layer that stays alive underneath your work, and remembers.

~/workspace — session
$ claude
/hello            # loads context, recaps last session
> remember: we chose Postgres over Mongo — audit log needs joins
   saved to memory/decisions.md
/bye              # captures lessons, persists memory

# days later — a brand-new session, zero re-explaining
/hello
> why did we pick Postgres?
   recall: audit log needed relational integrity · memory/decisions.md
$ git clone https://github.com/MDGrey33/claude_boilerplate.git ~/src/claude_boilerplate
View on GitHub

Ships today as claude_boilerplate — the Memnyx rename is in progress. Requires the Claude Code CLI; markdown memory works with zero extra setup.

Boilerplate copies and diverges.
Memnyx stays alive underneath your work.

Without it

Every session starts cold. You re-explain the stack, the decisions, the gotchas — and the context dies when the window closes.

With Memnyx

Context loads at /hello, lessons persist at /bye, and the next session already knows what you decided and why.

Three vertebrae

What the spine is made of.

Layered memory

Markdown files give fast, deterministic, always-available memory — with an optional semantic backend (cognee knowledge graphs or a Wikibase provenance graph) layered on for retrieval. Works with nothing extra; degrades gracefully when no backend is set.

Session lifecycle

/hello loads context, checks MCP health, and recaps where you left off. /bye summarizes the work, captures lessons, and persists memory — so no session ever starts from nothing.

Lessons that evolve skills

Wrap-up captures mistakes, conventions, and patterns; /lessons feeds /skills-manager to add, update, and review skills. /contribute flows generalized, scrubbed lessons back into the boilerplate — it gets sharper as you use it.

Three commands in

Up and remembering in a minute.

Clone the spine, oncegit clone …/claude_boilerplate.git ~/src/claude_boilerplate — outside your workspace.
Initialize your workspace/setup-workspace init --workspace ~/workspace deploys skills, generates CLAUDE.md, bootstraps identity.
Then every session/hello brings context in, /bye takes lessons out. Semantic memory is optional, later: /setup-cognee or /setup-wikibase.
Open by default

Yours to read, fork, and keep.

Apache-2.0 licensed Built in the open No signup, no SaaS

An open-source project by Roland Abou Younes & Sotiris Michalakos.

Give Claude Code a memory.

One clone, one init, and every session after this one remembers the last.

$ git clone https://github.com/MDGrey33/claude_boilerplate.git ~/src/claude_boilerplate
Star on GitHub