ntrp 0.3.4

Personal AI assistant with persistent memory and multi-source knowledge integration

ntrp

ntrp is entropy – the measure of disorder in a system. Your calendar, emails, notes, browser tabs, half-remembered conversations – it all accumulates. This project exists to reduce it.

I built this for myself. ADHD and scattered attention meant I kept losing track of things (e.g. what I said, what I planned, what I was supposed to follow up on). So I made an assistant that hooks into my stuff and actually remembers.

What it does

• Persistent memory: learns facts and patterns across conversations, consolidates them over time
• Scheduled tasks: morning briefings, daily reviews, health tracking – runs autonomously on a schedule
• Connected sources: Obsidian vault, Gmail, Google Calendar, browser history, web search (so far)
• Shell access: runs commands, manages files, sends emails
• Any LLM: Claude, GPT, Gemini built-in; OpenRouter, Ollama, vLLM, or any OpenAI-compatible endpoint via custom models

Memory

Schedules

Connections

Install

uv tool install ntrp    # backend (or: pip install ntrp)
bun install -g ntrp-cli # terminal UI (or: npx ntrp-cli)

Quick start

Create ~/.ntrp/.env with at least one LLM provider key and model variables (see .env.example):

mkdir -p ~/.ntrp
cat > ~/.ntrp/.env << 'EOF'
ANTHROPIC_API_KEY=sk-ant-...
NTRP_CHAT_MODEL=claude-sonnet-4-6
NTRP_MEMORY_MODEL=claude-sonnet-4-6
NTRP_EMBEDDING_MODEL=text-embedding-3-small
EOF

ntrp serve   # starts backend, prints a one-time API key
ntrp         # terminal UI (separate terminal) – paste the key on first launch

The server generates an API key on first run and prints it once. Enter it in the UI setup screen — it's saved to your system keychain. Use ntrp serve --reset-key to regenerate.

See docs/setup.md for Google OAuth, Telegram, Obsidian, custom models, Docker, and other integrations.

Requirements

• Python 3.13+, uv
• Bun (terminal UI)
• At least one LLM provider API key

Inspired by

• opencode – terminal UI
• letta – persistent memory and personalized approach
• hindsight – graph memory structure

Versioning

Pre-1.0: 0.minor.patch. Minor bumps for new features (may be breaking), patch bumps for fixes. All changes go through PRs, except docs and minor non-code updates.

License

MIT