lolabot

Your AI Chief of Staff — Personal Assistant framework for Claude Code. Email, semantic memory, task management, content security.

Why Lola?

Lola is a batteries-included AI Chief of Staff. She handles email, remembers everything, manages your tasks, and keeps your files organized. Clone the repo, run setup.sh, and point Claude Code at it — she's ready in minutes.

Lola runs standalone with just Claude Code. Optionally, deploy her on AI Maestro for persistent memory across sessions, inter-agent messaging via AMP, and multi-machine orchestration.

What You Get

Email Client — IMAP/SMTP with multi-account support, SPF/DKIM/DMARC verification
Semantic Memory — Hybrid Memvid + SQLite for fast search over facts, events, learnings, decisions
File Index — Track files across local, remote, and cloud locations (OneDrive, Google Drive, S3, iCloud)
Task Management — Eisenhower Matrix prioritization system
Content Security — 34-pattern prompt injection defense, email sanitizer, attachment risk assessment
Skills — File processing, mail handling, memory delegation
HEIC/Image Processing — Convert Apple image formats for AI consumption

Quick Start

Paste this into Claude Code and you are done:

Set me up with lolabot, a personal-assistant framework.

1. Create a folder at ~/assistant and cd into it
2. Clone https://github.com/23blocks-OS/lolabot.git into ~/lolabot
3. Run ~/lolabot/setup.sh ~/assistant and answer its questions using sensible
   defaults — ask me only for my name and what I want to call you
4. Copy lolabot.yaml.example to lolabot.yaml in ~/assistant
5. Tell me the single command I need to run next, and stop

Do not configure email or anything requiring passwords. I will do that later.

Full instructions, a launcher for one-word startup, and troubleshooting: INSTALL.md.

What happens on first run

Your assistant introduces itself and asks four questions — what to call you, what its job is, what
it should handle every week, and how you will know in six weeks whether it was worth it. It writes
your answers to brain/charter.md and into its own instructions, then gets to work.

Every question can be skipped. Open with real work instead and it does that first.

Optional: For persistent memory, inter-agent messaging, and multi-machine orchestration, deploy Lola on AI Maestro.

# Clone the repo
git clone https://github.com/23blocks-OS/lolabot.git

# Scaffold a new PA instance
./lolabot/setup.sh ~/my-assistant

# Configure your instance
cd ~/my-assistant
cp lolabot.yaml.example lolabot.yaml
# Edit lolabot.yaml with your settings

# Set environment
export LOLABOT_HOME=~/my-assistant

# Start Claude Code in your instance directory
cd ~/my-assistant && claude

How It Works

Clone & scaffold — setup.sh creates your PA instance directory with config, tools, and skills
Configure — Edit lolabot.yaml with your email accounts, preferences, and paths
Launch — Run claude in your instance directory — Lola has everything she needs
Scale up (optional) — Deploy on AI Maestro to add more agents, messaging, and orchestration

Lola is a framework, not a product. Fork it, customize CLAUDE.TEMPLATE.md, swap tools in and out, and make her yours.

Want more agents beyond Lola? Browse 150+ specialist personalities at The Agent Library and give them skills with the Plugin Builder.

Project Structure

your-assistant/          # Your PA instance (private, never pushed)
├── CLAUDE.md            # Generated from template
├── lolabot.yaml         # Your config (git-ignored)
├── tools/               # From lolabot (symlinked or copied)
├── skills/              # From lolabot
├── brain/               # Your tasks, notes, company info
├── memory/              # Your journals, profiles, goals
├── emails/              # Your cached emails
└── indexes/             # Your search indexes

Tools

Tool	Purpose
`tools/email_client.py`	Read, send, reply, forward, search emails
`tools/email_sanitizer.py`	Content security for inbound email
`tools/memory_indexer.py`	Semantic memory with Memvid + SQLite
`tools/file_indexer.py`	Multi-location file discovery and search
`tools/heic-convert.sh`	Apple HEIC to JPEG conversion
`tools/memory-integrity-check.sh`	Detect unauthorized file modifications

Skills

Skill	Purpose
`file-processing`	Document classification, organization, indexing
`mail-handler`	Safe email interaction rules and trust model
`pa-memory-delegation`	User memory management (facts, events, learnings)
`pa-onboarding`	First-run charter — name, role, standing work, how success is measured

Configuration

All configuration lives in lolabot.yaml. See lolabot.yaml.example for all options.

The single required environment variable is LOLABOT_HOME, pointing to your PA instance directory.

Security

lolabot includes defense-in-depth for AI assistants:

Prompt injection scanning — 34 patterns detected in inbound email
Trust model — Operator (verified sender) vs. external content separation
Email authentication — SPF/DKIM/DMARC verification
Content wrapping — External content tagged as data-only, never executed
Attachment risk assessment — Dangerous file types blocked
File integrity monitoring — SHA256 checksums on critical files

Requirements

Claude Code (Claude Max or API key)
AI Maestro (optional — persistent memory, messaging, multi-machine)
Python 3.10+
uv (recommended) or pip
Python packages: memvid-sdk, pyyaml, pillow-heif

License

MIT — 23blocks Inc.

Works standalone or on AI Maestro — the OS for AI-first organizations.