Hermes Agent as a Farmer's AI Advisor — Built on Android,...

Hermes Agent Challenge Build Submission

What I Built

A working agricultural AI advisor powered by Hermes Agent, running entirely on an Android phone via Termux, tested with real farming questions from Silchar, Assam, Northeast India.

No laptop. No GPU. No cloud server. Just a phone, Termux, and Hermes Agent doing real agentic work for farmers who cannot afford enterprise AI tools.

The Problem

Farmers in Cachar district, Assam ask questions like:

My rice leaves have brown spots near the edges. What disease is this?
When should I plant Boro rice and which variety should I choose?
Give me a 3-month farming plan starting from December.

These are not questions a static chatbot answers well. They need an agent that reasons across multiple steps, applies local agricultural knowledge, and gives actionable, region-specific advice.

Hermes Agent does exactly this.

Setup — Android + Termux + Hermes Agent

My environment:

Device: Android phone
Terminal: Termux → proot-distro Ubuntu (ARM64)
No laptop, no desktop, no GPU

Installation was a single command:

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

Hermes Agent auto-detected my ARM64 Termux environment and installed cleanly. This matters — it means the same setup works for any developer in rural India running Termux.

Model used: Google Gemini via OAuth (free tier) — zero cost, no API key purchase needed.

The Demonstrations

Demo 1: Rice Disease Diagnosis

My prompt to Hermes Agent:

I am a farmer in Silchar, Assam. My rice leaves have brown spots near the edges. What disease is this and how do I treat it?

Hermes Agent's response (summarised):

Hermes correctly identified two candidate diseases based on the symptom description:

Brown Spot (Fungal) — oval/circular spots with yellow halo, common in low-fertility soils. Treatment: Propiconazole 25% EC (1ml/litre) or Mancozeb 75% WP (2-2.5g/litre), plus balanced NPK fertilization.

Bacterial Leaf Blight — wavy brown edges spreading from leaf tips. Treatment: Streptocycline + Copper Oxychloride. Avoid excess Urea. Drain standing water.

Hermes then gave Silchar-specific advice: visit the local Krishi Vigyan Kendra (KVK) in Cachar with a leaf sample before purchasing chemicals.

This is not generic advice. It is regionally grounded, actionable, and honest about when a farmer should seek local expert confirmation.

Demo 2: Boro Rice Planting Calendar

My prompt:

What is the best time to plant Boro rice in Cachar district, and what variety should I choose?

Hermes Agent's response (summarised):

Nursery sowing: Mid-November to Mid-December
Transplanting: Mid-December to Mid-January
Harvest: April to May (before monsoon)

Recommended varieties specific to Assam: Joymati, Bishnu Prasad, Jyotiprasad, Kanaklata, IR-36, IR-50 — with maturity timelines and cold-tolerance tips for Silchar winters.

Sources for seeds: Assam Seeds Corporation, local block agriculture offices, KVK Cachar.

Demo 3: 3-Month Farming Plan

My prompt:

Create a simple 3-month farming plan for a rice farmer in Silchar starting from December

Hermes Agent's response:

Hermes generated a complete week-by-week plan across December, January, and February — covering seedbed preparation, cold protection, transplanting spacing (20x15cm), water management, fertilizer application timing, weed control, and pest monitoring. It even flagged the pro tip of draining water before Urea application to prevent runoff.

This is multi-step planning — not a single lookup but a structured, sequenced agricultural roadmap.

Why This Matters

Hermes Agent ran these three demonstrations on a phone in Termux. The same phone a farmer's son in Cachar district carries. The same environment that costs nothing beyond internet access.

The agentic capabilities that made this work:

Tool chaining — Hermes reasoned across multiple considerations (climate, soil, disease vectors, local institutions) in a single response without being explicitly prompted to do so.

Regional grounding — It applied Assam-specific knowledge (KVK Cachar, Assam Seeds Corporation, local variety names) rather than giving generic pan-India advice.

Honest uncertainty — It told the farmer to get a physical leaf sample confirmed before buying chemicals. An agent that knows the limits of its own remote diagnosis is more trustworthy than one that does not.

Planning capability — The 3-month farming plan was a genuinely structured multi-step output, not a list of generic tips.

Connection to My Broader Work

I have spent the past year building the Divine Earthly ASI system — a sovereign, offline-first agricultural AI for rural Indian farmers, running on quantized 0.5B models via llama.cpp on ARM64 Termux. My system is designed for zero-connectivity environments.

Hermes Agent occupies a different but complementary space: it requires internet connectivity but brings mature agentic infrastructure (skill learning, session memory, 28 tools) that would take years to build from scratch.

What they share: the belief that capable AI should run where farmers are, not just where servers are.

What Hermes Agent Does That Impressed Me

The skill-learning loop. After complex sessions, Hermes distills experience into reusable skill files. For agricultural use, this means the more farmers use it, the better it gets at local, regional questions — accumulating what I call Samskara: the impression that evolves future behavior.

See my companion Write submission for a deeper exploration of this Vedic parallel:
Samskara and the Self-Improving Agent

Reproduce This Yourself

Requirements: Android phone with Termux, or any Linux machine.

# Install proot-distro ubuntu (Termux only)
pkg install proot-distro
proot-distro install ubuntu
proot-distro login ubuntu

# Install Hermes Agent
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

# Set up free model (Google Gemini OAuth)
hermes model
# Select: Google Gemini via OAuth + Code Assist (free tier)

# Start
hermes

Then ask it a farming question. It works.

推荐订阅源

DEV Community