Inspiration
It is surprisingly hard to answer a simple question at the end of a busy day: What did I actually do, and what mattered? Existing activity trackers can reconstruct time, but often at the cost of sending sensitive window titles, notes, and behavioral data to a cloud service. Echo started from a different premise: memory assistance should be useful and private by default.
What it does
Echo is a local-first Windows activity journal. It records foreground application activity on-device, merges it into readable sessions, and presents a daily timeline alongside notes, moods, and nearby moments. A user can select a point in time and replay the surrounding context to reconstruct what they were doing without searching through raw logs.
Privacy controls are part of the core experience: recording can be paused, applications can be excluded, idle-time and retention preferences are configurable, and data can be backed up or exported by the user. Echo also supports Simplified Chinese and English without adding clutter to the Memory interface.
An optional GPT-5.6 reflection mode is off by default. Before every request, Echo builds a minimized, sanitized preview and asks for explicit confirmation. Window titles, file paths, URLs, email addresses, photos, and raw activity records are never included. Notes remain excluded unless the user separately opts in. The API key is stored in Windows Credential Manager.
How we built it
Echo is a native Windows desktop application built with Python, PySide6, SQLite, and pywin32. Codex and GPT-5.6 supported implementation, debugging, interface iteration, privacy review, automated test design, and documentation throughout Build Week.
Echo existed before the submission period, so the repository clearly documents the work added during Build Week. Meaningful extensions include memory replay, redesigned timeline and settings experiences, session merging, bilingual UI switching, optional privacy-reviewed AI reflections, secure key storage, recorder recovery, midnight splitting, export integrity, and an isolated synthetic demo mode.
The --demo mode is important for safe judging: it creates a complete synthetic day in a temporary directory, pauses normal recording, never reads the user's real Echo database, and makes no API request unless a tester explicitly configures it and confirms the sanitized preview.
Challenges
The hardest challenge was making an AI feature compatible with a product whose main promise is privacy. We treated sanitization, consent, and data minimization as product behavior rather than a policy document. Another challenge was keeping a desktop recorder reliable across crashes, idle periods, application switches, and activity that crosses midnight while keeping the interface responsive.
Accomplishments
- A useful memory-reconstruction flow without cloud-first tracking
- Privacy tests proving sensitive identifiers and window titles do not enter AI payloads
- Secure API-key storage through Windows Credential Manager
- Synthetic demo isolation for judges and reviewers
- Local backup and portable export with SQLite integrity
- Bilingual UI switching without disrupting the existing Memory layout
- A responsive background AI workflow that does not freeze the desktop interface
What we learned
Privacy becomes much stronger when it is visible and testable. Showing the exact sanitized payload before a request gives the user meaningful control, while automated tests turn privacy claims into engineering constraints. Codex was especially valuable for tracing cross-cutting behavior across recording, storage, UI state, and tests, and for turning design decisions into repeatable validation.
What's next
Next, Echo will improve local search and reflection quality, expand configurable redaction, add stronger onboarding for privacy controls, and make the demo/test build easier for judges to run without rebuilding from source. The long-term goal is a calm, trustworthy memory layer that remains owned by the person whose life it records.
Log in or sign up for Devpost to join the conversation.