Audtheia

An open-source ecological monitoring system that combines AI-powered species detection, automated environmental analysis, and research-grade data records into a single pipeline. Point a camera at a habitat, and Audtheia handles the science from detection to final report.

Research-grade monitoring, without a full research team on-site

Many of the most important ecosystems are also the hardest to monitor continuously. Field sites may be remote, difficult to access, or require around-the-clock observation that simply isn't practical with limited personnel. Audtheia was built to close that gap. A camera, an internet connection, and the free Audtheia setup are enough to run a rigorous, structured monitoring pipeline from anywhere.

Your data stays entirely within platforms you own and control. Credentials never leave your device or your N8N instance. The complete pipeline, including detection models, analysis workflow files, database schemas, and proxy source code, is published under the MIT license so anyone can use it, adapt it, or build on it without restriction.

Four connected components, one continuous pipeline

Detection happens at the camera. Analysis happens automatically. Data lands in your database. Reports generate on a schedule. Each component lives on its own platform, communicates through simple connections, and can be individually adapted to your target species and ecosystem.

01 / Roboflow

Species Detection

Your live video feed is processed at 60 FPS inside a Roboflow Workflow. When a species appears in frame, the detection is immediately passed through four custom processing blocks: one draws the live visual overlay, one converts the detection into structured data, one sends a quick AI-powered environmental note to the record, and one dispatches the full detection to the analysis pipeline. The reference implementation ships with a marine sponge classifier trained on 126 species. Any compatible detection model can be swapped in for different target species or ecosystems.

RF-DETR RTSP · 60 FPS Custom models
02 / N8N RTSP Analyst

Nine-Agent Analysis

Each detection triggers nine AI agents that work through the observation systematically. One agent identifies the taxonomy and seasonal timing. Another checks historical sightings of the same individual. A third validates the GPS location and elevation. Others query conservation databases, pull real-time environmental conditions, assess habitat quality, and generate a satellite map of the site. When all nine finish, the results are written to your Airtable database as a single 72-field observation record.

9 agents 72 data points GBIF · IUCN · NOAA
03 / N8N Daily Reporter

Automated PDF Reports

On a configurable schedule (daily by default), five agents read all recent observations and synthesize them into a structured research report. The report covers species diversity, environmental and climate patterns, habitat health, satellite spatial analysis, and a FAIR-compliant data summary. The finished PDF is saved to your Airtable base and available for download from the researcher dashboard.

5 agents Scheduled PDF FAIR synthesis
04 / Dashboard

Researcher Interface

A browser-based interface that connects directly to your Airtable base using your own credentials. You enter your Airtable token and base identifiers once, and the dashboard displays your species observations, satellite imagery, and PDF reports. No Audtheia account is needed. Your credentials are stored only in your own browser and sent only to Airtable's servers.

Airtable REST Local credentials No account needed

Built for any field environment

Audtheia works in marine, terrestrial, estuarine, and freshwater habitats. The RTSP Analyst includes separate climate agents for marine and terrestrial environments that can each be turned on or off independently, so you only gather data that is relevant to your specific site.

Marine

Open water, reef systems, intertidal zones, and pelagic environments. Sea surface temperature, salinity, ocean currents, and depth data from NOAA and Open-Meteo Marine.

Terrestrial

Forest, grassland, shrubland, and upland sites. Vegetation index, soil composition, biome classification, and atmospheric conditions from SOILGRIDS and Open-Meteo.

Estuarine

Mixed marine and freshwater transitional zones. Both the marine and terrestrial climate agents can be enabled at the same time for full environmental coverage.

Freshwater

Rivers, lakes, and inland aquatic systems. Elevation, watershed context, and atmospheric data provide the environmental baseline for freshwater habitat characterization.

The satellite mapping agent generates Mapbox imagery at three zoom levels and is optional. Turning it on only when spatial data is needed for a project keeps API usage proportional to the value it adds.

Every observation built for open science

FAIR data principles (Findable, Accessible, Interoperable, Reusable) are a widely used standard in ecology and biodiversity research. Every Audtheia observation is structured to meet all four, making records compatible with established biodiversity databases and archiving workflows from day one.

F

Findable

Each detection receives a unique Observation ID. Records include rich taxonomic metadata sourced from GBIF, iNaturalist, and IUCN, making every observation self-describing and searchable within the database.

A

Accessible

All data is retrievable via the Airtable REST API using your own credentials. CSV export is available at any time directly from Airtable. No Audtheia intermediary is required for data access.

I

Interoperable

Taxonomic fields follow Darwin Core conventions. Timestamps use ISO 8601. Coordinates are recorded in WGS 84. Schema documentation enables integration with other biodiversity and conservation data systems.

R

Reusable

Complete schema documentation and annotated example CSV files are in the repository. Every record includes a computed data quality score and provenance metadata documenting which data sources and agents contributed to it.

A transparent quality score for every observation

Every observation receives a quality score from 0 to 100. The score combines four weighted components and is stored alongside the record so researchers can filter their dataset based on their own quality thresholds for analysis.

Score components and their weights:

Data completeness 40% API success rate 30% Detection confidence 20% Validation checks 10% ───────────────────────── Quality Score 0 – 100

Research-grade threshold: ≥ 85. Observations below this threshold are kept in the database but flagged in the quality field, letting researchers apply their own filtering criteria.

Cite this work

Audtheia is archived on Zenodo as a versioned, citable release. The archive includes trained models, workflow files, Airtable schemas, proxy source code, and field deployment documentation.

Portalatin, A. (2025). Audtheia Environmental Monitoring System. DOI: 10.5281/zenodo.17842738. MIT License. https://github.com/AudtheiaOfficial/audtheia-environmental-monitoring
DOI: 10.5281/zenodo.17842738

Andy Portalatin

Andy Portalatin is a marine biologist with an M.S. in Biology concentrated in marine sciences. Audtheia grew directly from his graduate research. The official Porifera classifier included in the repository, trained on 16,915 labeled images across 126 tropical marine sponge species, was produced as part of his thesis work and serves as the reference implementation for the detection layer.

The goal behind Audtheia is broader than any single research project. Monitoring ecosystems rigorously has historically required either significant infrastructure or sustained physical presence. The full pipeline is published under the MIT license and documented so that anyone with a monitoring objective can use it: research agencies building species baselines, university programs teaching ecology or computer vision with a real-world system, independent researchers studying sites that are hard to access, and students or community scientists who want to contribute structured observational data to the scientific record.

Audtheia is a living project. Any deployment in a real field context will reveal edge cases, missing configurations, and ecosystem-specific adaptations that make the system better. Collaborations, curriculum integrations, and proposed extensions are all welcome. If you are developing a monitoring program, designing a course that could use a complete real-world pipeline, or have ideas that would improve Audtheia's coverage, opening a proposal on the GitHub repository is the right place to start.

Open a proposal on GitHub