Agentic AI

Overview

---

---

• If a file is .gitignored, create an empty placeholder file in the corresponding location to preserve the directory structure.

./
├─ backend/
│   ├─ main.py
│   ├─ autonomous_agent.py
│   ├─ task_generator.py
│   ├─ agent.py
│   ├─ memory.py
│   ├─ file_processor.py
│   ├─ notifier.py
│   ├─ config.py
│   ├─ tasks.json
│   ├─ uploads/        # Create empty
│   ├─ results/        # Create empty
│   ├─ .env            # Create empty
│   ├─ tasks.json      # Will be generated with first commands below
│   ├─ lab_loop.lock   # Auto-generated by server
│   └─ agentic_lab.log # Auto-generated by server
├─ frontend/
│   └─ ai-lab-dashboard/
│       ├─ package.json
│       ├─ vite.config.ts
│       └─ src/
│           ├─ Dashboard.tsx
│           ├─ TaskMonitor.tsx
│           ├─ MemoryMetrics.tsx
│           └─ KnowledgeGraph.tsx
└─ README.md

This package provides a fully autonomous AI research lab with:

• Multi-agent parallel task execution
• Self-generating tasks with prioritization
• File ingestion & persistent memory
• Cross-referencing and insight generation
• Email / Slack / Console notifications
• Logging & metrics
• A live React dashboard featuring:
  • Task monitor
  • Memory metrics
  • Interactive knowledge graph

---

⚠️ Kill Switch Notice
UPDATE: Use pkill -9 -f uvicorn or pkill -9 -f python.
If the backend server goes rogue and respawns endlessly despite kill -9 <PIDs>, a kill switch is built into the autonomous loop.
Run this inside backend/ (where the server is running):

export STOP_AGENT=1

Then stop Python:

pkill -f python

Congratulations—you’ve saved the system!

---

Setup

First run ollama serve, update the model name in agent.py, or run ollama run gemma:2b in a separate shell.

Frontend

Node.js, React/Vite

Runs on localhost:5173

cd frontend
npm i
npm run dev

Backend

Python

Runs on localhost:8000

cd backend   # in a new terminal

# follow the .example files to populate .env and config.py
# leave tasks.json empty when initializing
touch tasks.json .env config.py
mkdir uploads results

pip install -r requirements.txt

echo "[]" > tasks.json

uvicorn main:app --reload

For reruns after initial setup:

cd backend

# uploads/ and results/ already exist
# Save past results/ and empty the dir for new runtime
# wipe tasks.json and initialize with empty list
python -c 'import json; json.dump([], open("tasks.json", "w"))'

uvicorn main:app --reload

Optional: clear all files and logs before restarting:

python -c 'import json; json.dump([], open("tasks.json", "w"))'
rm -f ./results/*

uvicorn main:app --reload

---

Open Source ❤️

Made with love, please use this with pure intent. If you improve this project, please lmk at TechNYC.io.