Ways to integrate.
Call a model. Get a result.
Every plan includes all validated models and capabilities. Each returns a real-time measurement and, where applicable, a prediction of when the state will change, with a confidence interval.
Cognitive Load
Cognitive Load refers to the amount of effort a person allocates to a task. It is recognized as an important factor in determining a person’s level of performance in a task and during daily life activities (Devos et al., 2020).
Eye tracking: Cognitive Load can be measured through physiological tools. Eye tracking is a sensitive, quantifiable, accurate real-time measure of cognitive effort via eye responses such as changes in the pupil (Fehringer, 2021), blinking (Chen, Epps, Ruiz, Chen, 2011), and other eye movements (Irwin & Thomas, 2010; Kramer, 1991; Van Orden, Limbert et al., 2001; Greef et al., 2009; Klinger, Tversky et al., 2010).
Assumes the same activity continues without intervention.
Predicted future level: Low, Moderate, or High
Time to reach the future state: seconds
Probability (%): for the current level and other levels
Confidence interval (0–1, low–high): for each output
Attention
Attention refers to how a person directs and distributes their focus during a task or activity (Nideffer, 1976). The width of a person’s attentional focus, from narrow to broad, plays a critical role in task performance. The ability to match attentional width to the demands of a situation is a key driver of success across a wide range of activities (Nideffer, 1990; Nideffer & Sagal, 2006; Wulf, 2013).
Eye tracking: Attention can be measured through eye movements. Eye tracking provides a real-time, objective window into how attention is being deployed, whether a person is locked onto a specific target or actively scanning a wide field. The HarmonEyes Attention solution captures attentional width through fixation patterns, saccade frequency and duration, and gaze distribution across the visual field.
Optimal performance depends on matching attentional width to the task at hand. Mismatches (a narrow focus during a high-awareness task, or a broad focus during a precision task) are associated with reduced accuracy, slower reactions, and increased error rates.
Mental Readiness
Mental Readiness is a real-time, session-level metric derived from cumulative cognitive load and grounded in the mental fatigue literature, which defines mental fatigue as a decrease in mental performance resulting from cognitive overload (due to task duration and/or workload), independent of sleepiness.
It reflects how much cognitive capacity a person has left in reserve, as shaped by their cognitive load over time. People rarely operate at a single, steady level of mental effort. Cognitive load naturally fluctuates, and periods of sustained high cognitive load gradually deplete available resources (Borragán et al., 2017; Hu & Lodewijks, 2020). When mental readiness runs low, the effects are tangible: attention narrows, decision-making slows, and errors become more likely. Tracking how much of a session is spent at high cognitive load is therefore a key indicator of sustained performance, resilience under pressure, and the likelihood of cognitive overload (Boksem & Tops, 2008; Chen et al., 2018).
A running percentage of time spent in each demand state, updated every second starting one minute into a session:
For example, a session showing 70% Low / 30% High indicates that most of the session was spent with meaningful cognitive reserve intact.
Fatigue
Fatigue is a state of increased situational sleepiness associated with reduced alertness, slower reactions, and a greater tendency to fall asleep (Hu and Lodewijks, 2020). It impairs focus and responsiveness, especially during prolonged or repetitive tasks. Fatigue is driven by a combination of sleep debt (how long it has been since a person last slept) and the body’s internal 24-hour clock. Sleep is the most effective way to relieve fatigue.
Eye tracking: Changes in blink behavior are especially indicative of increasing fatigue (Barbato et al., 1995; Caffier et al., 2003; Schleicher et al., 2008; Cori et al., 2019; Abe, 2023).
Predicted future level: Alert, Somewhat Fatigued, Fatigued
Time to reach the future state: seconds
Confidence interval (0–1, low–high): for each output
A ready-made front end for Theia output.
An optional display layer for teams who need a front end but don’t want to build one. Configurable widgets size to your canvas and render live capability output in real time. No UI work required.
Live display
Real-time Cognitive Load and Fatigue readouts, with discrete or continuous scales and configurable widget sizing on the canvas.
In-session controls
Thresholding, refining, baselining, annotation, and prompting, all available live during a session.
Replay & review
After a session, replay results with every interaction, add annotations, and track task type.
Export & resources
Save, upload, and download session data as CSV. Built-in access to tutorials, updates, and dashboard info.
Any camera, plus dedicated eye-tracking hardware.
Runs on the hardware you already have.
The Theia SDK turns standard eye-tracking hardware into human-state insight at a fixed per-frame compute cost. No GPU, no cloud round-trip, no calibration. Figures below reflect validated production targets.
Download the full spec sheet (PDF) →Release notes
Every Theia SDK and Theia Dashboard release, newest first.
Version 2.4
August 2026 LatestFatigue model v2.4
An updated fatigue model reports on three levels instead of four. “Neither Alert nor Fatigued” and “Rather Fatigued” are consolidated into “Somewhat Fatigued.”
Version 2.3
July 2026Browser-based dashboard
The dashboard runs in a web browser with no desktop install. Every user gets a private session.
- “Mental Workload” is now “Cognitive Load,” and “Drowsy” is now “Fatigued”
- The Mental Readiness gauge tracks changes in load more closely
- Webcam tracking: smoother live gaze with fewer gaps and more reliable camera startup on Windows
- Scene and webcam recordings survive an unexpected stop, so footage is no longer lost
- Restyled update notifications, with a single press to restart and apply
- First-run license prompt, a clickable license status indicator, and automatic entitlement refresh on startup
- Export session predictions to Excel (.xlsx), with a ledger of every recorded value
- macOS auto-update is architecture-aware; an Apple-Silicon install no longer updates to an Intel build, or vice versa
- No more backend crash after very long session runtimes
- Mental readiness no longer reports “No predictions” in the 1-minute window
- Sessions survive a page refresh, and Start Session disables while starting to prevent double-starts
Version 1.1.0
May 2026Mental readiness indicator
A real-time mental readiness widget, with a 1-minute, 1-hour, and all-time lookback.
Attention indicator
Real-time attention classification (focused vs. distributed).
- More time window options for the cognitive load history
- View is now its own top-level menu, separating widget visibility from configuration
- Submenus open as right-flyouts and stay open until you click outside
Version 1.0.0
February 2026 Initial releaseThe first release: the Theia Dashboard for real-time cognitive load and drowsiness monitoring, and the Theia Python SDK for programmatic integration.
Predates the 2.3 rename: “Mental Workload” became “Cognitive Load,” “Drowsiness” became “Fatigue.” Original wording kept for the record.
- Live monitoring: real-time cognitive load from a generalized base model, plus four-level drowsiness detection (Alert, Neither Alert nor Drowsy, Rather Drowsy, Drowsy) on a color-coded bar
- Cognitive load graph: discrete categorical levels or a continuous scale, over a 3, 5, or 10-minute window
- Scene camera: live video from the connected eye tracker
- Sessions: start and stop from the footer and annotate live
- Layout & settings: resizable panels, per-panel visibility toggles, license management, and header status indicators
- Real-time streaming: start and stop sessions and view included solutions
- Requirements: Python 3.12 and Pupil Labs Neon (
PL) or the mock tracker (WT) for development
Bringing human-state intelligence to 8 billion devices.