Skip to content
Best for UCSC

HydroClawNics

Hydroclawnics is an autonomous AI agent that manages a hydroponic farm, constantly monitoring pods across temperature, pH, EC, humidity, water levels, etc, then taking corrective action in real time.

Code Demo Video Devpost

Core trick

Supervisor agent compares each pod's live pH, EC, and temperature against species-specific ranges, then calls actuator functions for fans, heaters, or dosing.

Tags

agentsrobotics/hardware

Built with

arduinofastapinemoclawnemotronnemotron-3-supernvidianvidia-brewopenclawreactthree.jswebsocket

Description

Modern hydroponic farms generate constant sensor data across hundreds of pods... but existing tools just display it. They don't act, explain, or learn. Hydroclawnics changes that. We built an autonomous agent powered by NVIDIA NemoClaw and Nemotron 3 Super that reads live sensor data from every pod, reasons about crop health using species-specific target ranges for lettuce, tomato, basil, and spinach, and calls physical control functions to keep plants alive. Whether that be fan speed, heater, cooler, humidifier, or pH dosing, our agent's actions can be the defining moment between a healthy plant and one that needs to be discarded. Every decision streams to a live reasoning feed so operators understand exactly why the agent acted. The system runs on a hybrid architecture: one physical pot with real Arduino sensors, and 200 simulated pods running as a drop-in replacement. Flipping a single environment variable switches between hardware and simulation — the agent never knows the difference. Nemotron 3 Super's 1M token context window means the agent can supervise 500+ pods in a single reasoning cycle (the scale of a real commercial greenhouse) while only activating 12B parameters per token for fast, efficient inference on Brev. With multiple subagents, this could possibly be expanded further. Built for three Hack-A-Claw tracks: Cloud Track - full deployment on NVIDIA Brev with CI/CD pipeline NemoClaw Bonus - agent orchestration and security sandbox via OpenClaw Santa Cruz - Hydroponics is currently the most human-reliant field of agriculture, and bringing autonomous agents to monitor plants for people massively reduces the cost of mistakes and timeliness. Tech stack: NemoClaw / OpenClaw · Nemotron 3 Super · NVIDIA Brev · FastAPI · React · Three.js · WebSocket · Arduino
GalexY727/Hydroclawnics★ 20 forksJavaScriptMITCreated May 16, 2026 · last push Sep 2, 2026

README excerpt

# Hydroclawnics

Hydroclawnics is a real-time hydroponics farm simulation and agent dashboard. It demonstrates simulated grow pods, WebSocket updates, table-level agents, supervisor logic, corrective tool calls, action logs, and a frontend feed that explains what the system is doing.

The default demo is local-first and free-hostable: it does not require NVIDIA Brev, NVIDIA hosted inference, GPUs, Ollama, or paid cloud inference.

## Architecture Overview

```text
FastAPI backend
  ├─ SimulatorEngine writes pod sensor snapshots
  ├─ WebSocket endpoint streams pod and agent updates
  ├─ Table agents evaluate crop-specific pod ranges
  ├─ Supervisor synthesizes farm-wide status and directives
  └─ SQLite/message bus/action logs preserve agent activity

React frontend
  ├─ 3D farm view and pod detail UI
  ├─ WebSocket-fed pod updates
  └─ Agent reasoning/action feed
```

LLM behavior is selected with `LLM_PROVIDER`:

- `none`: deterministic offline demo, safest for public hosting.
- `ollama`: local OpenAI-compatible Ollama endpoint.
- `nvidia`: optional NVIDIA-hosted compatibility mode.
- `openai`: generic OpenAI-compatible hosted mode.

## Run Locally: Deterministic Demo

From the repository root:

```bash
python -m venv .venv
source .venv/bin/activate
pip install -r hydroclawnics/requirements.txt
cp .env.example .env
LLM_PROVIDER=none DEMO_MODE=true DEMO_AUTOSTART_AGENTS=true uvicorn hydroclawnics.main:app --reload
```

On Windows PowerShell:

```powershell
python -m venv .ven

Team