Pomona
Open edge AI for safer greenhouse and hydroponic decisions.
Connect simulated or real sensor data to a local pipeline with MQTT, compact reasoners, deterministic safety checks, human-reviewed suggestions, and an operational dashboard.
From sensor reading to guarded suggestion¶
Sensor / Simulator → MQTT → Pomona Core → SQLite → Model Router
→ specialist reasoners → Safety Checker → Automation Engine → Dashboard
Safety boundary: the LLM advises; it never directly controls actuators. Deterministic checks and human approval remain in the control path.
Explore Pomona¶
-
Try without installing
Test guarded tomato, irrigation, and nutrient scenarios in the public browser demo.
-
Run the local stack
Start MQTT, the APIs, safety services, automation suggestions, and the dashboard with Docker Compose.
-
Understand the safety design
See how model output is constrained by schemas, deterministic rules, and explicit approval.
-
Use specialist reasoners
Browse Pomona's small-model catalog and the lifecycle status of each published or local candidate.
Pomona ecosystem¶
Each public surface has one clear job. Use the documentation as the main hub, then move to the code, models, or live API without losing context.
-
Documentation
Guides, architecture, safety boundaries, project status, and roadmap.
-
Source and community
Platform source, Docker quickstart, issues, discussions, and contributions.
-
Models, data, and demo
Published reasoners, the greenhouse dataset, collection, and browser demo.
-
Live automation API
Explore the suggestion-only automation service and its guarded API endpoints.
What works today¶
| Capability | Current state |
|---|---|
| Sensor ingestion | MQTT and REST ingestion with SQLite persistence |
| Dashboard | Read-only sensor, guarded-pipeline, audit, and service views |
| Reasoning | Deterministic rules plus guarded local-model routes |
| Safety | Dedicated deterministic safety checker and actuator-command gate |
| Automation | YAML rules produce suggestions with manual approve/reject decisions |
| Public demo | Static Hugging Face Space with preset greenhouse scenarios |
Pomona is an early MVP, intended for development and evaluation—not unattended production control. See the current project status for verified capabilities and remaining work.
Start in five minutes¶
In a second terminal:
Read the complete installation guide
Models and data¶
Model weights and published datasets stay on Hugging Face; platform code, schemas, routing, and deterministic safety logic stay on GitHub.
- Pomona model and dataset collection
- Tomato risk reasoner
- Water and irrigation reasoner
- Greenhouse sensor dataset
Help build the open agriculture stack¶
Pomona welcomes focused contributions to documentation, tests, dashboards, simulators, device integrations, and guarded reasoning. Start with the project status, then choose a small issue or open a discussion.