Skip to content

zscole/gru

Repository files navigation

Gru

CI License: MIT Python 3.10+

Your personal army. Spin up autonomous agents that complete tasks, learn your patterns, and remember everything.

You: Research the top 3 competitors and build me a comparison landing page
Gru: Spinning up research agent... Found 3 competitors. Analyzing features.
     Spinning up dev agent... Building landing page.
     Done. Deployed to vercel: https://compare.yoursite.com

You: What did I discuss with Sarah last week?
Gru: 3 interactions found:
     - API redesign planning (Tuesday)
     - Q1 budget review (Wednesday)
     - Deployment timeline sync (Friday)

You: Book Nobu for Saturday 7pm, venmo @mike for the concert, and remind me Friday
Gru: Done:
     - Reserved Nobu for Saturday 7:00 PM (party of 2)
     - Sent $50 to @mike via Venmo
     - Reminder set for Friday

Quick Start

1. Get your keys (5 min)

Key Where
Anthropic API Key console.anthropic.com
Telegram Bot Token Message @BotFather > /newbot
Your Telegram ID Message @userinfobot

2. Deploy (1 min)

Deploy on Railway

Set these env vars:

  • ANTHROPIC_API_KEY
  • GRU_TELEGRAM_TOKEN
  • GRU_ADMIN_IDS

3. Chat

Open Telegram, message your bot:

Build me a landing page for a coffee shop

That's it.

What Gru Can Do

Multi-Agent Orchestration

Gru spins up specialized agents that work in parallel to complete complex tasks.

Say this Gru does this
"Research AI developments and build me a PoC" Spawns research agent + dev agent
"Organize my inbox and summarize the important stuff" Email agent processes, filters, summarizes
"Build a landing page based on this screenshot" Vision agent analyzes, dev agent builds
"Check my calendar and prep me for tomorrow's meetings" Calendar agent + research agent on attendees

Agents learn from each task and share context, getting better over time.

Autonomous Actions

Gru doesn't just tell you how - it does it for you.

Say this Gru does this
"Book Nobu for Saturday 7pm" Makes the OpenTable reservation
"Venmo @sarah $30 for lunch" Sends the payment
"Order my usual from Chipotle" Places the DoorDash order
"Schedule a call with Mike tomorrow 3pm" Creates the calendar event
"Email John the project update" Sends the email

Personal Knowledge Graph

Gru remembers every conversation, email, and meeting - and connects the dots.

Ask this Gru answers
"What did I discuss with Sarah last month?" Pulls all interactions with timestamps
"Who's working on the API redesign?" Maps relationships from your conversations
"What topics have I been focused on?" Analyzes patterns across all sources
"When did I last talk to the investors?" Searches across email, Slack, calendar

Persistent Memory

Gru learns as it goes. Every interaction builds context that makes it more useful.

  • Pattern recognition - "You usually check metrics on Fridays"
  • Preference learning - Remembers your usual orders, common contacts, meeting preferences
  • Cross-session context - Picks up where you left off, even weeks later
  • Proactive insights - "You mentioned following up with Mike - no response yet"

Proactive Intelligence

Gru anticipates what you need before you ask.

  • 6am daily briefing - Calendar, unread emails, Slack mentions
  • Smart reminders - Surfaces uncommitted follow-ups and deadlines
  • AI research - Monitors HN/Reddit, builds working PoCs of interesting projects
  • Anticipation - "Traffic is heavy, leave now for your 4pm"

Self-Healing

Gru monitors itself and fixes problems automatically.

  • Self-diagnostics - Continuous health monitoring
  • Stuck detection - Notices when agents aren't progressing
  • Auto-recovery - Restarts failed services, fixes configuration
  • Bug detection - Identifies errors and spawns fix agents
  • Performance monitoring - Tracks slow operations and resource usage

Multimodel Support

Gru routes tasks to the best model for the job.

Task Type Routed To
Complex reasoning Claude Opus
Code generation Claude Sonnet
Simple cleanup Ollama (local)
Web search Ollama (local)
Vision tasks Claude (required)

Run models locally with Ollama for cost savings and unrestricted capabilities:

GRU_OLLAMA_URL=http://localhost:11434
GRU_OLLAMA_MODEL=llama3:8b

Development Tools

The original Gru - still the best way to code from your phone.

  • Natural language - "Build me a landing page for a coffee shop"
  • Screenshots - Send a UI mockup, get working code
  • Voice notes - Speak your request while walking
  • Ralph loops - Iterative AI development cycles
  • Live deploy - Ship to Vercel in one message

Platform Setup

Telegram (recommended)
  1. Message @BotFather > /newbot
  2. Copy the token
  3. Get your user ID from @userinfobot
GRU_TELEGRAM_TOKEN=your-token
GRU_ADMIN_IDS=your-user-id

Full guide

Discord
  1. Create app at discord.com/developers
  2. Get bot token, enable Message Content Intent
  3. Invite with bot + applications.commands scopes
GRU_DISCORD_TOKEN=your-token
GRU_DISCORD_ADMIN_IDS=your-user-id

Full guide

Slack
  1. Create app at api.slack.com/apps
  2. Enable Socket Mode, get both tokens
  3. Add /gru slash command
GRU_SLACK_BOT_TOKEN=xoxb-xxx
GRU_SLACK_APP_TOKEN=xapp-xxx
GRU_SLACK_ADMIN_IDS=U01ABC123DE

Full guide

Self-Hosted Installation

Standard
git clone https://github.com/zscole/gru.git
cd gru
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env  # Edit with your keys
PYTHONPATH=src python -m gru.main
Docker
git clone https://github.com/zscole/gru.git
cd gru
cp .env.example .env  # Edit with your keys
docker-compose up -d

Documentation

Doc Description
Multi-Agent Orchestration Spawning and coordinating agents
Self-Healing Diagnostics, auto-recovery, bug detection
Autonomous Actions Book reservations, send payments, order food
Knowledge Graph Query your relationships and history
Proactive Intelligence Morning briefings, pattern learning
Commands Full command reference
Ralph Loops Iterative AI development cycles
Configuration All environment variables
Troubleshooting Common issues and fixes
MCP Plugins Extend with MCP servers

Why Gru?

Other AI assistants answer one question at a time, then forget.

Gru is your personal army. It spins up agents to handle complex tasks, learns your patterns over time, remembers every conversation, and takes real action in the world.

Feature ChatGPT Siri Gru
Multi-agent orchestration No No Yes
Autonomous task completion No No Yes
Self-healing No No Yes - auto-recovery
Book restaurants No Limited Yes - OpenTable, Resy
Send payments No No Yes - Venmo
Order food No No Yes - DoorDash
Query past conversations No No Yes - full history
Learn your patterns No No Yes
Persistent memory Limited No Yes - cross-session
Morning briefings No No Yes
Run code No No Yes
Deploy apps No No Yes

Architecture

Telegram/Discord/Slack
         |
         v
   Orchestrator
         |
         +---> Provider Router
         |         |
         |         +---> Claude API (complex tasks, code, vision)
         |         +---> Ollama (local models, simple tasks)
         |
         +---> Agent Pool (spawn, coordinate, parallelize)
         |         |
         |         +---> Research Agents
         |         +---> Dev Agents
         |         +---> Task Agents
         |
         +---> Knowledge Graph (entities, relationships, history)
         +---> Memory Store (facts, preferences, learned patterns)
         +---> Proactive Engine (triggers, anticipation, briefings)
         +---> Autonomous Actions (reservations, payments, orders)
         +---> Connectors (Gmail, Calendar, Slack)
         +---> MCP Plugins (extensible tools)

Privacy & Security

Your data stays yours. Gru runs on your infrastructure - not ours.

  • All data stored locally in SQLite
  • Knowledge graph lives on your machine
  • No telemetry, no tracking, no cloud dependency
  • Conversation history never leaves your server
  • Browser sessions stored in your home directory

Security controls:

  • Admin allowlist - only approved users can interact
  • Action confirmations - payments and reservations require approval
  • Supervised mode - approve commands before execution
  • Workspace isolation - sandbox agent file access

Development

pip install -e ".[dev]"
PYTHONPATH=src pytest tests/ -v
ruff check src/ tests/
mypy src/

Funding & Sponsors

This project is funded by the $GRU token community on Solana.

Contract Address: S782cLXpcS5agE3T6g7ADZ8LkmQsiDMKd7S2GzfGapp

Buy $GRU: Apps.fun | Jupiter Exchange

Special thanks to our sponsors and community members who support open-source AI development.

License

MIT

About

Self-hosted AI agent orchestration service controlled via Telegram, Discord, or Slack.

Resources

License

Contributing

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors