Open-source · FAIR data · Marine + Terrestrial

Environmental monitoring at the intersection of computer vision and multi-agent AI

Audtheia watches your field site so you don't have to be there. Aim your camera at a designated habitat, and the system automatically identifies species, cross-references them against global biodiversity databases, records 72 data points per sighting, and generates a structured research report, all without manual intervention and fully customizable.

60 FPS RTSP video inference
72 data points per observation
9 ecological analysis agents
14+ integrated databases
RF-DETR · YOLOv11 · Custom Classifiers Multi-species detection · Roboflow Workflow · RTSP · 60 FPS

From camera feed to research record, automatically

Set up a live field camera on a research site or use pre-recorded videos. Audtheia calls different cloud platform services to determine what species are visible, obtain environmental data, store everything in a structured database, and generate daily PDF reports. Each step hands off cleanly to the next with no manual work required at any stage.

01

Roboflow

official-porifera-classifier-ju8er/12

This is where species identification happens. Roboflow is used to run computer vision models through Real-Time Streaming Protocol (RTSP) videos. Audtheia is designed to analyze live detections from a custom or pre-trained model inside your Roboflow Workflow. When a species appears, the workflow converts the raw detection into structured data, draws the live overlay on your video feed, sends the detection to AI agents for deeper analysis, and generates a quick preliminary environmental note using an AI language model. The reference Porifera classifier covers 126 marine sponge species. You can swap in any detection model for your own target species.

  • Hosted inference endpoint at detect.roboflow.com
  • Works with live RTSP streams, webcams, or uploaded video files
  • Four custom blocks handle video overlay, data conversion, AI analysis, and webhook dispatch
RF-DETR Custom classifier RTSP
02

N8N

RTSP Analyst Workflow · Daily Reporter Workflow

N8N is the analysis engine. Every time Roboflow detects a species, N8N receives the detection and runs nine specialized AI agents in sequence. Each agent collects different data points per observation: where exactly is this species found, what is its taxonomic classification, is it an endangered species, what are the current environmental conditions? A second workflow, the Daily Reporter, runs on a customizable schedule and compiles everything into a PDF research report.

  • 9 agents in the RTSP Analyst, covering taxonomy, GIS, biodiversity, climate, and mapping
  • 5 agents in the Daily Reporter for scheduled PDF generation
  • All API keys and credentials stay inside your N8N instance, never transmitted to any client
Multi-agent Automation REST
03

Airtable

api.airtable.com · REST · browser CORS

Airtable is used as the main database. Each observation lands in your own Airtable base with 72 structured data points covering species identity, location, environmental conditions, conservation status, habitat quality, and more. You own this database entirely. The Audtheia website dashboard reads directly from it using your own credentials, with no intermediary involved.

  • Species Observations table with 62 columns per record
  • Environmental Mapping table with satellite imagery links
  • Daily Reports table storing finalized PDF download links
FAIR data REST API Structured
+

Integrations

14+ scientific data sources

All nine analysis agents retrieve environmental information from various established databases (GBIF, iNaturalist, SoilGrids, etc.), so each observation record arrives already annotated with conservation status, taxonomy, habitat data, and satellite imagery, without you having to query any of them manually.

  • GBIF · iNaturalist · IUCN · NOAA · Open-Meteo
  • Seven additional scientific and environmental databases
  • Mapbox satellite imagery at 10×, 13×, 15× zoom for every detected location
GBIF iNaturalist IUCN NOAA Mapbox Open-Meteo +9 more

Your accounts, your data

Audtheia works on a bring-your-own-account model. You create accounts on the platforms listed below, connect them together following the setup guide, and your data stays entirely within systems you own and control. Nothing passes through this website. Most platforms here have free tiers that cover everything you need to run a full monitoring deployment.

API key Roboflow Trains and runs your species detection model. This is where your camera feed is processed. roboflow.com →
N8N Runs the analysis workflows. Nine AI agents process each detection and produce structured research data. n8n.io →
API key Airtable Stores all your observation records. The dashboard reads directly from your own Airtable base. airtable.com →
API key Anthropic Powers the preliminary environmental analysis that runs inside the Roboflow detection workflow. anthropic.com →
API key OpenAI Powers the N8N analysis agents. Any compatible language model provider can be used instead. platform.openai.com →
Free tier Render Hosts the small proxy server that lets the demo page send images to Roboflow from a browser. render.com →
API key Mapbox Generates satellite imagery of each detection location at three zoom levels for spatial context. mapbox.com →
API key API Template Converts the Daily Reporter's analysis into a downloadable PDF research report. apitemplate.io →
Free, open GBIF Global Biodiversity Information Facility, species occurrence records and taxonomy data. No key required. gbif.org →
Free, open iNaturalist Community-sourced biodiversity database with species sightings contributed by researchers worldwide. inaturalist.org →
API key IUCN Red List Global conservation status database. Used to flag threatened, endangered, or critically endangered species. iucnredlist.org →
Free, open NOAA Oceanographic and climate datasets for environmental context in marine and coastal deployments. noaa.gov →
Free, open Open-Meteo Open-source weather and atmospheric data, no registration needed. Used for real-time site conditions. open-meteo.com →
Free tier API Ninjas Species and biodiversity data used for taxonomic enrichment and ecological cross-referencing. api-ninjas.com →
Free, open OpenTopoData Open elevation data used to characterize habitat and geospatial context for each observation. opentopodata.org →
Free, open OpenStreetMap Open geographic database used via the Overpass API for location context and habitat mapping. openstreetmap.org →
Free, open SoilGrids Global soil data from ISRIC, used for substrate and soil composition context in terrestrial deployments. soilgrids.org →

Open science, archived and citable

Audtheia and its reference dataset are freely available, permanently archived, and citable. The full codebase is on GitHub under the MIT license. The official Porifera classifier dataset is published on Roboflow Universe under CC BY 4.0. Both can be used, adapted, and built upon without restriction.

Zenodo archive
DOI: 10.5281/zenodo.17842738

Complete system archive including trained models, workflow configurations, Airtable schemas, proxy source code, and field deployment documentation. MIT licensed.

10.5281/zenodo.17842738
Source code
AudtheiaOfficial / audtheia-environmental-monitoring

Includes the proxy source code, N8N workflow files, Airtable schemas, Roboflow workflow configurations, and full documentation. Contributions welcome via pull request.

Roboflow Universe
rf Marine Sciences Research Station / official-porifera-classifier-ju8er

Reference dataset for Porifera detection: 16,915 labeled images across 126 marine sponge species. Version 12 trained with RF-DETR (Medium). The starting point for Audtheia marine deployments. CC BY 4.0 licensed.

official-porifera-classifier-ju8er/12