Inside Timetrack: keep the work visible
A concrete engineering walkthrough: preserve short manual entries, explain parallel AI sessions, and keep exported reports available for review.
By Nicholas Howell · Full Stack Technologies ·
Start with the detail a reader could lose
A fifteen-minute manual entry can matter even when automated sessions run for hours. Timetrack records AI coding sessions and manual activity, then turns those records into views and reports. One engineering decision is to preserve the short entry when preparing a summary, rather than take only the longest sessions.
A repeatable check uses ten long agent sessions and one fifteen-minute manual entry. All eleven remain in the summary's starting data. The test checks that the summary retains those sample records; their duration alone does not establish what the work achieved.
For a founder building an app, this is a useful review question: does the transformation preserve the information a person needs to make a decision? A clear screen can still hide an important record.
Define what the recorded time means
Concurrent AI sessions keep their own durations. Two sessions running during the same hour can add up to more than one hour of session time. The dashboard makes overlap visible so the reader can distinguish recorded activity from elapsed human time.
Manual entries carry a separate source label. Session attribution connects the selected project to that project's client, keeping those two labels consistent. The labels explain where the records came from and what they belong to; session duration alone is not a measure of human effort.
Keep an exported report available for review
Export stores a snapshot of the report document. Opening that saved report retrieves the stored document instead of recalculating it against today's records and rules. A fresh draft can reflect later changes without silently replacing the saved document.
This is a stored export snapshot, not an immutable archive: saved reports can be deleted. The broader lesson is to decide whether a historical view should show what was recorded then or what the system would calculate now.
Separate AI-assisted development from AI product features
Nicholas used AI assistance while building Timetrack, including work on session attribution and reconciliation. Deciding what a report should preserve and checking that behavior still require engineering judgment.
The product also has separate, optional model features, including title generation and report narration. A summary can start from recorded labels with blank descriptions and next steps, without inventing a narrative. Generated prose still needs review: a model's sentence is not independent evidence that an outcome occurred.
Describe where the data actually goes
Timetrack is self-hosted across configured services, rather than entirely on-device. Local hooks collect activity within allowed repository roots; selected metadata goes to a Supabase collector and is shown in a Vercel-hosted dashboard. Default hook capture discards prompt and tool contents, while paths are shortened or aliased rather than removed entirely.
Optional title generation is off by default. When explicitly enabled, it can send bounded excerpts of human messages to a model provider. Report narration is another optional model feature. Deciding which features to enable requires understanding their inputs and destinations, not relying on a broad privacy label.
Explore the interface, then apply the questions to your app
The public Timetrack page includes an interactive dashboard with sample data: switch between day and week, then inspect a point on the dial. The other views illustrate the interface. That demonstration is separate from the capture, authentication, and export systems described here; clicking through it does not exercise those integrations.
For your app, choose one consequential result and trace it back to its inputs. Decide what each label means, which records must survive a summary, and what a saved document preserves. A small, repeatable check can make the next product decision much clearer.
- Explore Timetrack's sample-data interface
- How Nick works through these decisions with you
- Review your AI prototype before launch
- Work through a real task together