Show HN: Dayflow – A git log for your day

Original Article ↗ Hacker News Discussion ↗

Comparison to Windows Recall / similar tools

  • Several commenters see this as similar to Recall/Rewind since both continuously capture the screen.
  • Others stress a key difference: Recall is about later retrieval, whereas this focuses on semantic summarization of time.
  • Some argue that once you have screen data, there’s nothing stopping a system from doing both retrieval and summarization.

Privacy, security, and deployment model

  • Strong concern about sending sensitive on‑screen data (banking, passwords, work) to cloud models like Gemini.
  • Many appreciate the open‑source, self‑hostable design and local‑only mode; this is contrasted with Microsoft or third‑party hosted tools.
  • Some note that enterprise Gemini projects can avoid training on user data, but trust in large vendors remains shaky.
  • A few users are uneasy that a new GitHub account ships software that could be spyware, but others say the source is available to inspect.

Local vs cloud AI: quality, cost, and resources

  • Reported quality gap: Gemini 2.5 Pro ≈ “A‑level”; local Qwen 2.5 VL ≈ “B–/C+”.
  • Local models work via Ollama/LM Studio etc., but are CPU/GPU intensive and drain laptop battery; suggestion to only process while plugged in.
  • Gemini costs are significant: ~1M input tokens per hour of video, but current free tier covers typical personal use.

Use cases and target users

  • Popular ideas:
    • Reconstructing billable hours for lawyers, contractors, and freelancers (automatic, granular time logs).
    • Helping people with ADHD or procrastination understand distraction patterns and task flow.
    • Generating standup summaries and “what did I do yesterday?” reports for engineers.
  • Some imagine pairing this with speech‑to‑text, calendar tools, and automation to execute tasks from natural language.

Workplace surveillance & legal concerns

  • Significant worry that employers could use such tools for invasive monitoring, turning it into “dystopian” productivity policing.
  • Commenters distinguish voluntary self‑tracking from boss‑imposed tracking.
  • Legal concerns raised about recording video calls (e.g., Zoom) in all‑party‑consent jurisdictions; unclear how laws treat 1 fps continuous capture.

Technical behavior and performance

  • The app records at 1 fps in 15‑second chunks, then analyzes ~900 frames every 15 minutes; some question whether this is truly “lightweight.”
  • Users report periodic CPU spikes/heat during local processing, and one person estimates ~€1/hour in cloud spend without careful configuration.
  • Multi‑monitor behavior: current approach records the focused display; this is seen as a pragmatic 90/10 solution but misses context on secondary screens (e.g., a video call while working elsewhere).

Platform support, integrations, and extensibility

  • macOS‑only for now; several people ask for Linux and Windows versions.
  • Strong interest in integrating other data sources: wearables/HealthKit, phone logs, custom apps.
  • Suggestions to provide an API / plugin system so others can extend it, possibly with an “App Store”‑like ecosystem.
  • Ideas to improve efficiency: pause capture on idle, during fullscreen media, or based on power‑adapter status.

Trust, UX, and related tools

  • Many praise the UX, onboarding wizard, copywriting, and clear privacy explanations.
  • Some want faster initial feedback (immediate first card) and better debugging tools (screenshot tests, clearer error surfacing).
  • Related or alternative tools mentioned: ActivityWatch, ScreenMemory, screenpipe, CLI‑based window trackers, and text‑only flows like doing.
  • A few users note naming confusion with an unrelated “Dayflow” and question the “git log” metaphor, seeing the UI as more calendar‑like than terminal‑style.