HumanAgentWiki — AI skill for Claude Code

HumanAgentWiki

A local, private **"second brain" your AI agents can actually read** — semantic search + [MCP](https://modelcontextprotocol.io) over your own Markdown notes.

**Status: early WIP** — being extracted and generalized from a working private system. See [ROADMAP.md](ROADMAP.md).

---

What it is

HumanAgentWiki turns *your own notes* (plain Markdown) into a shared, searchable knowledge base that any **MCP-compatible AI agent** can plug into. Everything runs **locally on your machine** — your knowledge never leaves your computer.

Why

• 🧠 LLMs don't know your context. HumanAgentWiki gives them retrieval over what you actually wrote and think (RAG), so answers are grounded in your knowledge — not generic web fluff, and far fewer hallucinations.
• 🔒 Local & private. PostgreSQL + pgvector on your own machine. No cloud upload.
• 🌍 Multilingual. Uses BGE-M3 embeddings — excellent for Czech and 100+ languages, not just English.
• 🔌 Agent-native. Exposes an MCP server, so Claude and other MCP clients connect with zero glue code. Multiple agents share one brain.

Install

One line:

curl -fsSL https://raw.githubusercontent.com/petrludwig-collab/HumanAgentWiki/main/install.sh | bash

Or clone and run the installer:

git clone https://github.com/petrludwig-collab/HumanAgentWiki.git
cd HumanAgentWiki && ./install.sh

It checks prerequisites, creates a virtualenv, installs dependencies, brings up PostgreSQL (Docker if available, otherwise a local instance), writes `.env`, and creates the schema. Then use the `./haw` launcher:

./haw index     # index your notes (Markdown under ./notes)
./haw web       # web UI at http://127.0.0.1:8808
./haw serve     # MCP server for your agents

Try it instantly with the bundled demo: `NOTES_DIR=./sample_notes ./haw index && ./haw web`

How it works (plain English)

1. Your notes