💡 Agent Trust & Transparency Dashboard: Proactive Data Visibility as Product Feature

[github-actions[bot]]

Summary

Build a comprehensive trust dashboard showing users exactly what the agent can see, has done, and plans to do — with one-click data export and deletion. Position EU AI Act regulatory compliance as a product advantage rather than a checkbox, making TalkTerm the most transparent AI agent on the market.

Market Signal

EU AI Act transparency obligations became enforceable on August 2, 2026. Desktop AI apps that talk to users must disclose they are AI and show what data is processed — fines up to 15M EUR or 3% of worldwide turnover (Cooley, Goodwin law firm analyses). Ambient computing research shows trust-building features (activity indicators, data dashboards, easy deletion) are now table stakes for user adoption. Lapu AI's market positioning emphasizes "permissioned behavior" — asking before acting. Claude Desktop shows session costs but lacks comprehensive data visibility. No AI assistant offers a full trust dashboard with export/deletion controls.

User Signal

TalkTerm targets non-technical users who are inherently less comfortable with opaque AI behavior. The PRD mandates audit trail logging (FR27-29), confirm-plan patterns (FR20), and conversational error handling (FR34-35). Existing idea #420 (EU AI Act Transparency Mode) addresses regulatory disclosure but focuses on AI-nature disclosure, not the full trust UX. This proposal is complementary: #420 handles the "I am an AI" disclosure; this dashboard handles "here is everything I know about you and everything I've done."

Technical Opportunity

The audit trail infrastructure (FR27-29) provides the data foundation — every agent action is already logged with timestamp, type, and outcome. The OutputPanel's multi-mode display system (FR45) can host a new "Trust Dashboard" display mode alongside Document, ComparisonTable, and ActivityFeed. SQLite session storage captures action logs. This is incremental engineering on existing architecture: a new display mode + a data export endpoint + a deletion API. The three-zone layout naturally accommodates a data visibility panel.

Assessment

Dimension	Score	Rationale
Feasibility	high	Builds on existing audit trail + OutputPanel infrastructure; no new systems needed
Impact	high	EU AI Act compliance is legally required; trust is table stakes for non-technical user adoption
Urgency	high	EU AI Act enforcement began August 2, 2026 — this is no longer a future concern

Adversarial Review

Strongest objection: A trust dashboard could make TalkTerm feel "enterprise-heavy" and intimidate the non-technical users it aims to serve. Over-transparency might increase anxiety rather than trust.

Rebuttal: The dashboard is accessible from a settings/profile menu — not in the main conversation flow. The avatar announces it conversationally: "Want to see what I've been working with? Here's everything I know." The UX follows TalkTerm's core principle: complex information delivered through the avatar relationship, not raw data dumps. EU AI Act compliance is non-negotiable — better to make it a feature than a burden.

Suggested Next Step

Define the trust dashboard data model (files accessed, APIs called, tokens consumed, cost incurred, data sent to Anthropic). Map EU AI Act Article 52 transparency requirements to specific UI elements. Design the avatar-guided UX for accessing and narrating the dashboard. Create story in Epic 9 (Audit Trail) as an extension.