Agentic Vision-Language Models
A digital agronomist for every field.
DeepLeaf Agronomist is a large, agentic Vision-Language Model with a Mixture-of-Experts architecture. It sees the crop, reasons like a specialist, and acts — orchestrating narrow AI models and live field data through its own harness, flow engine, and the Data Hub MCP server.
The loop
See. Reason. Act. On repeat.
Not a classifier that returns a label — an agent that closes the loop from a photo in the field to an action grounded in that field's live data.
It reads the crop
DeepLeaf-Agronomist-1 reads symptoms and conditions straight from the image — training-free, visual-first, not a fixed class label.
It thinks like a specialist
An intent router sends the request through versioned flows; the MoE model reasons over the field's live data via MCP tool-use.
It answers and acts
It replies in voice and text and generates live UI — maps, charts, forms — with a legal, dose-correct recommendation.
It reads the crop
DeepLeaf-Agronomist-1 reads symptoms and conditions straight from the image — training-free, visual-first, not a fixed class label.
It thinks like a specialist
An intent router sends the request through versioned flows; the MoE model reasons over the field's live data via MCP tool-use.
It answers and acts
It replies in voice and text and generates live UI — maps, charts, forms — with a legal, dose-correct recommendation.
The farmer's journey
What the farmer does — and where DeepLeaf steps in.
From planning a season to selling the harvest, the loop closes at every step: the farmer acts, DeepLeaf gets ahead of the risk.
Plans the season: crop, variety and plot.
Agronomic plan
Crop and variety fit · Sowing window and density · Base fertilization plan · From soil, history and climate
Sows, and registers the plot.
Monitoring switched on
Satellite, weather, soil moisture · Live from day one
Feeds and irrigates, following the plan.
Input plans per zone
Irrigation + fertigation schedule · Fertigation and fertilization · Irrigation scheduling · VRA maps for per-zone dosing
Gets an alert, and walks only the flagged zones.
Proactive, from space
Forecasts stress and risk · Generates the scouting zones · Alerts him first
Scouts the zone, sends a photo or voice note.
Diagnosis and prescription
Diagnosis in seconds · Product, dose, timing, window · Agronomist-reviewed · VRA map, affected zones only
Harvests, sells, needs credit or cover.
Yield and provenance
Yield estimation and forecast · Provable field record · For the buyer, bank or insurer
Capabilities
One agent, the whole workflow.
Reads, not just classifies
A training-free, visual-first approach driven by image embeddings reads symptoms and adapts to new problems in real time, with no retraining.
Agentic MoE reasoning
A Mixture-of-Experts Vision-Language Model that reasons over the scene and calls tools through the DeepLeaf harness.
Versioned flow engine
An intent detector and router send each request through versioned flows built for many real-world scenarios.
Orchestrates narrow models
Computer-vision models (detection, segmentation, tracking, counting, digital-twin reconstruction, ChArUco 1mm, lane detection, quality/focus checks) and forecast models (climate, pest & disease, yield) under one orchestrator.
Live tool use over MCP
Grounded in the DeepLeaf Data Hub through its MCP server — real-time field data and actuators on demand.
Generative UI, not just chat
Beyond voice and text, it generates real-time UI/UX — maps, charts, forms — to enhance the conversation. Available in the SDK.
The model family
One family, two ways to build.
DeepLeaf-Agronomist-1-4B ships as an open-source Digital Public Good; the frontier MoE is hosted, with a full SDK.
Open Source · Digital Public Good · Vision-Language
DeepLeaf-Agronomist-1-4B
Compact, open model — build on it for free, on-device or self-hosted.
DeepLeaf-Agronomist-1
The frontier proprietary model — highest accuracy, hosted, with a full SDK for chat, real-time voice, and vision.
Notify me when DeepLeaf-Agronomist-1-4B ships
Open weights as a Digital Public Good — free for the world. Leave your email and we'll tell you the moment downloads open. Status: weights not released yet.
Coverage
Supported crops & conditions.
One model across the world's crops — reading pests, diseases, deficiencies, and stress from a single image, in the farmer's language.
Crops
Fungal diseases
Late blight · Powdery mildew · Downy mildew · Rusts · Botrytis
Bacterial & viral
Bacterial spot · Fire blight · Mosaic virus · Leaf curl
Pests & mites
Aphids · Whiteflies · Thrips · Spider mites · Armyworm
Nutrient deficiencies
Nitrogen · Potassium · Magnesium · Iron
Weeds & abiotic stress
Broadleaf weeds · Heat stress · Water stress · Frost damage
Who it's for
Built for the whole value chain.
Farmers & growers
Your own agronomist, in your pocket — 24/7, in your language.
Agronomists & field teams
Cover more fields and catch problems earlier, from your desk.
Developers
Build with the SDK — chat, real-time voice, vision, and generative UI.
How it compares
More crops. More languages.
Crops understood
Languages spoken
Directional comparison against common tool categories, not any specific product. Figures are approximate.
Build with DeepLeaf-Agronomist-1.
Start free on the open 4B model, or request access to the frontier model and SDK.