Claude is brilliant.But it forgets everythingyou talked about yesterday.Zé doesn't.
A persistent memory stack for your Claude Code. Vault, semantic search and knowledge graph working together so it never asks again what you already answered.
open-source · 3-minute install · works offline
How many times have you explained this?
Every new Claude conversation starts from zero. You re-explain the project. You re-paste the database schema. You re-state what you already decided about the architecture. You re-point the bug it already fixed last week. Time drains into repetition. So does patience. And Claude — who's genius — starts every day as the new intern.
Memory that survives.
Three components working in sync to turn your Obsidian vault into a persistent brain your Claude can query anytime.
Your notes are the source of truth
Markdown on your machine, versioned in git, offline, yours. Zé doesn't steal anything — it just organizes what you already have.
Semantic search over everything
Embeddings of every note via sqlite-vec. Ask in natural language and Zé pulls the right snippet, even if you wrote it with different words.
Connections only a graph can see
Local Neo4j storing projects, decisions, stories and pending items as connected nodes. Multi-hop questions RAG can't answer, the graph answers in milliseconds.
Zé has a CLI. And it's got personality.
After installing, you talk to Zé from the terminal. It remembers, pulls, queries the graph, and answers you right back.
Three minutes, one command, parrot in your terminal.
Requires docker, uv, git and the Claude Code CLI. If you have those, Zé handles the rest.