Dayflow turns screen activity into an automatic, AI-summarized git log of your day

Most time trackers answer a question nobody asked: which app was open. Two hours in Cursor can mean shipping a feature, debugging auth, reviewing a PR, or getting lost in setup, and a journal built on window titles cannot tell those apart. The honest version of the same record, written by hand or started with timers, fails because it depends on memory and discipline.

Dayflow, launched as a Show HN project under the tagline a git log for your day, removes the human from the recording step. It watches the screen in lightweight chunks, sends them to the user's chosen AI, and reconstructs the day as activity cards arranged in a timeline, with context-aware summaries of what was actually done. From that timeline it drafts standup updates with a GitHub-style activity grid, yesterday's highlights, today's priorities and blockers, and it adds a weekly review, distraction tracking, a chat interface over the journal, and Markdown export.

The privacy design is what makes a screen recorder tolerable. Recordings, timeline and database stay on the Mac by default; the user decides whether analysis runs on local models through Ollama or LM Studio or goes to a cloud provider through their own key or CLI, and storage limits purge old recordings automatically. The app is open source under MIT and installable with brew install --cask dayflow.

The September 2025 Show HN drew 480 points and 130 comments on Hacker News. By the time this batch catalogued the project, the repository showed 7,000 stars, 424 forks and 1,667 commits, with releases for macOS 14 and later.

Make the journal a byproduct of observation, then let AI name the activities

Screen content is the ground truth of work while app names are not: the pixels show whether a Cursor session shipped a feature or debugged auth, so the AI reads context, not window titles.

Capture is automatic, so the journal is complete without discipline; nobody has to start a timer, switch a task, or write anything down.

Local-first storage with an optional local-model pipeline answers the privacy objection that normally kills screen-recording products.

The summary layer meets users where they already are: a standup draft, a weekly review and a searchable chat replace the memory work without adding new habits.

Choosing the AI provider turns privacy, cost and quality into one user-controlled dial instead of a vendor lock-in.

For behavior that needs honest continuous recording, stop asking the user to log it: capture the evidence passively, summarize it with AI, and make privacy the default.

The September 2025 launch thread on Hacker News scored 480 points and 130 comments, and by the September 2026 catalogue the Dayflow repository had grown to 7,000 stars, 424 forks and 1,667 commits. The project ships as a macOS app with a Homebrew cask and remains MIT-licensed and local-first, with the stated workflow of capturing screen chunks, analyzing them with a chosen AI provider, and exporting the resulting timeline as Markdown for status updates, client notes and personal reviews.