nobrainer-memory

name	nobrainer-memory
description	Install memsearch persistent memory for Claude Code. Auto-captures every session as markdown notes, injects relevant context on every prompt. Uses local Ollama embeddings (nomic-embed-text) — no API key needed. Use when setting up a new machine or when user says "install memory", "setup memsearch", "nobrainer-memory", "dodaj pamiec do claude", "zainstaluj memory".

NoBrainer Memory — memsearch Installer

Installs memsearch persistent memory plugin for Claude Code. Philosophy: Markdown is the source of truth. Vector index is just a cache. Every session gets summarized into .md files. Semantic search injects relevant context automatically.

What Gets Installed

1. memsearch Python CLI (PyPI)
2. nomic-embed-text Ollama model — local embeddings, no API key needed
3. memsearch ccplugin registered in Claude Code plugins
4. Config: ~/.memsearch/config.toml (provider: ollama)

Memory files land in: ~/.memsearch/memory/YYYY-MM-DD.md (global, all projects)

Step 0 — Ask about memory scope

Before installing, ask the user:

Gdzie zapisywać memory?

1. Global (~/.memsearch/memory/) — jeden pool dla wszystkich projektów. Działa gdy zawsze otwierasz Claude z ~. Kontekst z różnych projektów miesza się w jednych plikach, ale semantic search i tak znajdzie właściwy.
2. Per-projekt (<project-dir>/.memsearch/memory/) — izolowana pamięć per repo. Wymaga otwierania Claude z folderu projektu (cd ~/Github/mój-projekt && claude).

Który tryb preferujesz? (1 = global, 2 = per-projekt)

Based on the answer:

• Global → proceed with default install, MEMORY_BASE=$HOME/.memsearch
• Per-projekt → ask which directory: W jakim folderze projektu? → set MEMORY_BASE=<project-dir>/.memsearch, note it in the summary

Store the choice in MEMORY_SCOPE variable for use in Step 8 summary.

Step 1 — Detect environment

python3 --version
which pip3 || which pip
which ollama || echo "OLLAMA_MISSING"
uname -s  # Darwin or Linux

If Ollama is missing:

• macOS: brew install ollama (if brew available) or instruct user to install from https://ollama.com
• Linux: curl -fsSL https://ollama.com/install.sh | sh
• If Ollama cannot be installed automatically, configure memsearch with openai provider instead (requires OPENAI_API_KEY)

Step 2 — Install memsearch CLI

Try in order until one succeeds:

# Option A: pip3 (preferred)
pip3 install memsearch

# Option B: if A fails with "externally-managed-environment" (macOS Homebrew Python)
pip3 install memsearch --break-system-packages

# Option C: pipx
pipx install memsearch

# Option D: uv
uv tool install memsearch

Verify: memsearch --version

Step 3 — Pull embedding model

# Start ollama server if not running (macOS)
# ollama is usually auto-started as a service

ollama pull nomic-embed-text

Verify: ollama list | grep nomic-embed-text

If Ollama unavailable, skip to Step 4 and set provider to openai.

Step 4 — Configure memsearch

memsearch config set embedding.provider ollama
memsearch config set embedding.model nomic-embed-text

If using OpenAI fallback:

memsearch config set embedding.provider openai
memsearch config set embedding.model text-embedding-3-small
# User must have OPENAI_API_KEY in their env

Verify: cat ~/.memsearch/config.toml

Step 5 — Install Claude Code plugin

Option A: Marketplace (preferred, simplest)

Use the Claude Code built-in marketplace commands:

# From Claude Code CLI:
marketplace add zilliztech/memsearch

# Or equivalently:
/plugin install memsearch

This handles downloading, registering, and configuring the plugin automatically.

Full docs: https://zilliztech.github.io/memsearch/claude-plugin/

Option B: Manual install (fallback if marketplace unavailable)

Check Claude Code plugins directory exists:

ls ~/.claude/plugins/installed_plugins.json

If it doesn't exist, Claude Code is not installed — tell the user to install Claude Code first.

Clone memsearch repo and copy plugin files:

TMPDIR=$(mktemp -d)
git clone --depth=1 https://github.com/zilliztech/memsearch.git "$TMPDIR/memsearch"

PLUGIN_VERSION=$(memsearch --version | sed 's/memsearch, version //')
PLUGIN_DIR="$HOME/.claude/plugins/cache/zilliztech/memsearch/$PLUGIN_VERSION"

mkdir -p "$PLUGIN_DIR/hooks" "$PLUGIN_DIR/skills/memory-recall" "$PLUGIN_DIR/scripts"
cp "$TMPDIR/memsearch/ccplugin/hooks/"* "$PLUGIN_DIR/hooks/"
cp "$TMPDIR/memsearch/ccplugin/scripts/"* "$PLUGIN_DIR/scripts/"
cp "$TMPDIR/memsearch/ccplugin/skills/memory-recall/"* "$PLUGIN_DIR/skills/memory-recall/"

rm -rf "$TMPDIR"

Step 6 — Register plugin (only for Option B manual install)

Skip this step if you used marketplace install (Option A) — it registers automatically.

Read ~/.claude/plugins/installed_plugins.json.

Check if memsearch@zilliztech already exists in the plugins object. If yes — update installPath and version. If no — add a new entry.

Add/update this entry in the plugins object (before the closing }):

"memsearch@zilliztech": [
  {
    "scope": "user",
    "installPath": "/Users/<USERNAME>/.claude/plugins/cache/zilliztech/memsearch/<VERSION>",
    "version": "<VERSION>",
    "installedAt": "<ISO_TIMESTAMP>",
    "lastUpdated": "<ISO_TIMESTAMP>"
  }
]

Replace <USERNAME> with $HOME resolved, <VERSION> with actual memsearch version, <ISO_TIMESTAMP> with current UTC time in ISO 8601 format.

IMPORTANT: Use Edit tool (not Write) to update the JSON. Preserve all existing plugin entries.

Step 7 — Verify installation

echo "=== memsearch ===" && memsearch --version
echo "=== config ===" && cat ~/.memsearch/config.toml
echo "=== ollama ===" && ollama list | grep nomic
echo "=== plugin ===" && ls ~/.claude/plugins/cache/zilliztech/memsearch/

All four checks should pass.

Step 8 — Initialize auto memory for current project

Claude Code has a built-in auto memory system (separate from memsearch) at: ~/.claude/projects/<project-hash>/memory/MEMORY.md

This file is automatically loaded into every conversation context for that project.

If the user is running this skill from within a project directory:

1. Detect the project memory path from the system prompt (look for "persistent auto memory directory at")
2. Create the memory/ directory if it doesn't exist
3. Create MEMORY.md with a basic template:

# Project Auto Memory

## Key Facts
- (add project-specific facts here)

## Conventions
- (add coding conventions, preferences)

## Detailed Topics
See `~/.memsearch/memory/` for semantic search memory.

4. Tell the user: "Auto memory initialized. Edit MEMORY.md to add project-specific context that should always be available."

Step 9 — Report to user

Print a summary:

memsearch installed successfully.

Version: <version>
Embeddings: ollama/nomic-embed-text (local, no API key needed)
Memory files: ~/.memsearch/memory/YYYY-MM-DD.md
Plugin: registered in Claude Code
Auto memory: <project memory path>/MEMORY.md (always loaded)

Restart Claude Code to activate. From next session:
  SessionStart  — injects last 2 days of notes as context
  Every prompt  — semantic search injects top-3 relevant memories
  Session end   — Claude Haiku summarizes session to .md
  Always        — MEMORY.md loaded into context (project-specific)

Error Handling

Problem	Fix
pip not found	Try pip3, pipx, uv in order
externally-managed-environment	Add --break-system-packages
Ollama not installed	Install via brew/curl or fall back to openai provider
ollama pull fails (no internet)	Use local provider: memsearch config set embedding.provider local
installed_plugins.json malformed	Read it, fix JSON, then edit
git not found	Download zip: curl -L https://github.com/zilliztech/memsearch/archive/main.zip -o /tmp/ms.zip && unzip /tmp/ms.zip -d /tmp/ms-extracted

Notes

• Memory is global (all projects share ~/.memsearch/memory/) when Claude Code is launched from ~
• If user always opens Claude Code from a specific project folder, memory will be per-project in <project>/.memsearch/memory/
• Haiku summarization uses the claude CLI — no extra setup needed if Claude Code is installed
• Milvus-lite (local .db) is the default backend — no Milvus server needed
• Watch process is skipped in lite mode — indexing happens once at SessionStart