Tether: Personalized ADHD Support Assistant

• Tether is a neurodiversity-aware support assistant that integrates LLM and retrieval-augmented generation to provide personalized interventions for software engineers with ADHD.
• It employs real-time desktop activity monitoring and gamification to tackle challenges like sustaining focus and initiating tasks in development workflows.
• The modular architecture ensures privacy and adaptability by combining native OS integrations with context-driven prompts for tailored, evidence-based support.

Tether: A Personalized Support Assistant is an LLM-powered desktop application developed to support software engineers with Attention Deficit Hyperactivity Disorder (ADHD) by delivering adaptive, context-aware assistance tailored to development workflows. Unlike generic productivity tools, Tether is engineered to mitigate core ADHD-related challenges—such as sustaining focus, initiating tasks, and self-regulating attention—by integrating local activity monitoring, retrieval-augmented generation (RAG), and gamification. The approach is grounded in engineering research methodology and targets real-time, personalized intervention in the software engineering context (Shah et al., 2 Sep 2025).

1. System Objectives and Core Functionality

The principal objective of Tether is to provide real-time, neurodiversity-aware support for software engineers with ADHD. Instead of functioning as a general-purpose productivity assistant, Tether is explicitly aligned with the practical and cognitive demands of SE workflows and ADHD symptomatology.

• Proactive engagement: Tether monitors desktop-level user activity—including application switch patterns, idle time, and active window tracking—through native OS integrations. Upon detecting sustained inactivity or potential distraction, it generates personalized notifications. These prompts are designed to be adaptive, varying their content and intensity according to recent user behavior.
• Contextualized conversational assistant: The chatbot ingests recent activity data and user-initiated queries to offer micro-interventions: breaking down tasks, facilitating motivational self-talk, and supplying emotional regulation strategies. The system is differentiated by its ability to access a repository of ADHD-specific literature and previously indexed interactions, enabling responses to be highly tailored.
• Retrieval-augmented generation (RAG): Instead of relying solely on autoregressive LLM completion, Tether employs a RAG pipeline that injects contextually relevant, evidence-based material into the prompt construction process for the LLM, ensuring semantic accuracy and individualization.
• Gamification features: Focus events, rapid recoveries from interruptions, and successful task initiations are rewarded via a points and badges system. This is designed both to reinforce positive work behaviors and to support self-regulation by providing immediate visual feedback.

2. Technical Architecture

The architecture of Tether is modular and designed to prioritize both privacy and extensibility.

Component	Implementation	Functionality
Activity monitoring	OS-level polling;	Tracks active window, idle periods, user inputs for context signal
	Flask backend +
	local SQLite
RAG pipeline	LangChain	Indexes ADHD-specific corpora, past user data, and queries Gemini
	+ Gemini (embed/gen)
Gamification engine	Electron-React UI	Renders progress, rewards, and achievement logs
Frontend	Electron + React	Chatbot UI, notifications, badge display
Backend	Flask REST API	Orchestrates monitoring, serves context to LLM and gamification UI

All major operations (activity tracking, prompt construction, badge display) occur locally, reducing privacy leakage. LLM calls (using Gemini) are selectively invoked for natural language generation tasks, with context embedded in prompts by serializing recent activity and relevant ADHD resource snippets.

3. User Interaction and Adaptation

Tether is designed for seamless integration into the daily workflow of software developers:

• Notification layer issues context-sensitive prompts during task drift or inactivity, leveraging historical activity patterns to adjust nudge frequency and content.
• The conversational chatbot allows users to initiate sessions at any time if they experience cognitive overload, require help with task decomposition, or need emotional regulation. The assistant can recommend granular strategies such as timeboxing or breaking tasks into atomic actionable units, tailored via context derived from monitoring signals and local history.
• Gamified progress tracking is displayed as a persistent sidebar. As users achieve concentration milestones (e.g., coding without task switches for a configurable period), badges are visually displayed and point tallies updated, reinforcing behavioral goals aligned with ADHD best-practice interventions.
• The interaction paradigm emphasizes non-disruptiveness, empathy, and high context-relevance, minimizing additional cognitive load for the user.

4. Validation and Preliminary Results

Tether’s design and internal assessment leverage engineering research methodology focused on utility and iterative improvement rather than formal summative evaluation at this stage.

• Comparative evaluation against extant ADHD support tools (e.g., browser blockers, generic chatbots, timetrackers) revealed that Tether uniquely provides developer-specific workflow integration, context-aware prompting, evidence-based RAG, and embedded gamification in a unified platform.
• Self-use validation by the development team indicated measurable improvements in contextual accuracy of responses, as determined by iterative cycles of prompt refinement and augmentation of the RAG module with additional ADHD literature and self-generated activity histories.
• No large-scale user studies have been conducted yet; performance metrics such as notification accuracy, engagement rates, and badge earn frequencies are tracked for future analysis. The absence of third-party user validation is identified as an area for further work.

5. Distinctive Features and Future Implications

Tether’s system design establishes a foundation for neurodiversity-aware support tools in SE and, more broadly, for professional domains where cognitive difference is under-served.

• Equity and inclusion: By supporting individualized attention, tailored micro-interventions, and positive reinforcement within the actual workflow of neurodivergent professionals, Tether aims to expand the accessibility of SE work environments.
• Technical extensibility: The modular architecture (with clear separation of monitoring, RAG, and feedback subsystems) makes integration of additional modalities—such as gaze tracking or environmental noise detection—feasible and under consideration.
• Generalizability: The LLM+RAG-based pipeline, which dynamically grounds assistance in both observed behavior and curated literature, is adaptable to other cognitive domains (e.g., ASD, dyslexia) and could be expanded for use cases outside software engineering.
• Planned enhancements: Incorporation of richer sensing (e.g., physiological signals for affect), formal user studies involving collaboration with healthcare professionals and neurodivergent engineers, and advanced reward dynamics in the gamification system are slated for future work.

6. Significance in the Context of Neurodiversity-Aware LLM Applications

Tether advances the field by specifying a comprehensive approach to real-time, personalized, AI-mediated support for software developers with ADHD:

• The system bridges local activity sensing with evidence-driven LLM responses.
• Retrieval-augmented prompt injection grounds dialogue in relevant, domain-specific information, mitigating the genericity that limits many LLM applications.
• Gamification targets self-efficacy and emotional regulation—both critical in managing ADHD—while minimizing stigmatization by embedding feedback naturally into the workflow.
• While not yet extensively evaluated by target users, the architecture and methodology set a precedent for future neurodiversity-aligned tool development in high-cognitive-load, technical professions.

A plausible implication is that such approaches, once validated through real-world user studies, could become a model for inclusive, intelligent, context- and condition-aware assistants across a spectrum of professional domains where individualized needs are not met by one-size-fits-all automation (Shah et al., 2 Sep 2025).