JerryZLiu/Dayflow

The automatic work journal. Privately turns your screen into a timeline of what you actually accomplished. Open-source and local-first.

Swift6049 starsPersonal AI & KnowledgeGitHub

Quality: solid 17/24

PAI: integrate 0.75

Overview

What it is — Dayflow is a native macOS app that silently captures periodic screen snapshots, analyzes them with a user-chosen AI provider, and assembles the day into a searchable, exportable work-journal timeline.
Problem — Knowledge workers lose accurate recall of what they actually did; manual time-trackers require discipline, and app-level logs tell you which window was open, not what the work was.
Who it's for — Mac-based professionals and developers who want a passive, privacy-respecting record of their workday without changing any habits or starting timers.
Notable — Full local-AI support via Ollama and LM Studio means zero data leaves the machine if desired; 6,049 stars in roughly eight months and v1.13 release cadence confirm it landed as the first serious entrant in passive screen-to-journal tooling.

Verdict

	Rating	Summary
Quality	solid (17/24)	Actively maintained Swift app with strong adoption and clear feature docs; held back only by absent dependency manifest and no visible test infrastructure.
PAI Relevance	integrate (0.75)	Fills PAI's complete blind spot in passive context accumulation — its Markdown timeline export is directly ingestible by a file-reading PAI skill at zero overlap with existing capabilities.

Quality Assessment

17/24 — actively-maintained / adequately-documented / solid

Health: 7/8 (actively-maintained)

Failed:

H8: FAIL — README contains no CI badge and no reference to .github/workflows/; CI status is unknown.

Passed:

H1: PASS — v1.13.1 release exists on GitHub Releases.
H2: PASS — Latest release dated 2026-05-26, less than 24 hours before appraisal.
H3: PASS — Last commit 2026-05-26, well within 6-month window.
H4: PASS — Last commit yesterday; push cadence is daily or near-daily.
H5: PASS — archived: false.
H6: PASS — 22 open issues: above zero (triaged) and well below 100 (not neglected).
H7: PASS — MIT license declared.

Documentation: 5/8 (adequately-documented)

Failed:

D4: FAIL — Code blocks present (Homebrew install, Build From Source) but neither appears under a heading titled Usage, Example, or Quickstart.
D5: FAIL — No section heading matching API, Configuration, Options, Reference, Commands, or Parameters.
D8: FAIL — No explicit Limitations, Caveats, Known Issues, or Trade-offs section; privacy notes appear inline under "Privacy" but no structured caveats heading.

Passed:

D1: PASS — README is present and populated.
D2: PASS — README is several thousand bytes with tables, GIFs, and multiple feature sections.
D3: PASS — Dedicated "Install" section covers both DMG download and brew install --cask dayflow.
D6: PASS — Opening paragraph states purpose within first 200 characters: "Dayflow understands the work you do on your Mac and turns it into a clear timeline of your day."
D7: PASS — https://dayflow.so is linked repeatedly as the canonical product/docs domain.

Engineering Signals: 5/8 (solid)

Failed:

E2: FAIL — Dependency manifest listed as "Not available"; no Package.swift or equivalent surfaced.
E3: FAIL — Cannot assess dependency count without manifest.
E4: FAIL — No test infrastructure mentioned in README, no manifest test script, no CI reference.

Passed:

E1: PASS — Swift is an explicitly typed compiled language.
E5: PASS — 6,049 stars far exceeds the 50-star threshold.
E6: PASS — ~8 months from creation to today; ~756 stars/month vs 2-star threshold.
E7: PASS — 329 forks exceeds the 5-fork threshold.
E8: PASS — Description is 97 characters and meaningfully describes the tool's purpose.

PAI Relevance

Dimension	Score	Assessment
Harvest Value	1	The passive screen-capture → AI-chunk-analysis → structured-timeline pipeline is an architectural pattern PAI has no equivalent of; the approach to chunking screen context for AI summarization is worth studying for any PAI context-accumulation subsystem.
Integration Readiness	1	Dayflow is a native Swift macOS app with no CLI or npm package; however its Markdown timeline export (configurable date range) lands at a known filesystem path, making a PAI file-reading skill a viable moderate-glue integration without wrapping the app binary.
Overlap Risk	0	No existing PAI skill, tool, or hook performs passive screen-activity capture or automatic timeline generation; ContextSearch and Knowledge operate on manually curated content, leaving this space entirely open.
Gap Fill	2	PAI's memory subsystem (WORK/, LEARNING/, KNOWLEDGE/) is fully manual — nothing feeds it without explicit principal action; Dayflow's Markdown exports would provide automatic ground-truth context about what the principal actually did, a clear functional gap with no existing coverage.

Composite: 0.75

What Next

Capture-to-Knowledge Pipeline (capture-layer spec): Configure Dayflow with Ollama as the local AI provider, run it through one full development session, then pull the exported timeline JSON from ~/.dayflow/. Feed that export directly into the C2K pipeline's design review as the canonical "what capture emits" specimen — the timestamped activity clusters, per-cluster AI summaries, and app-switch events become the concrete input schema the ingestion stage designs around. The abstract "what does capture output look like?" spec question closes with a real, parseable artifact rather than a whiteboard description.
Daily synthesis with fab: After each session, export Dayflow's timeline (File → Export or directly from ~/.dayflow/) and pass the plain-text version to fab with intent "daily standup synthesis". Fab selects the best-matching Fabric pattern and pipes the timeline through it, producing a structured standup or EOD reflection grounded in what actually appeared on screen rather than memory reconstruction. You also get ongoing signal on which Fabric patterns perform best against screen-sourced, AI-pre-annotated content — useful calibration data for fab's recommender as it sees a new content class.
PAI agentic session auditing: During active agent runs — skill iterations, eval loops, multi-step tasks — let Dayflow run passively with a short capture interval (30–60 seconds). After the session, diff the Dayflow timeline against what the agent reported doing. Discrepancies between screen-visible activity and agent-claimed actions surface dropped or hallucinated steps without adding any instrumentation inside the agent loop or modifying existing tooling.

Landscape Position

Category: Personal AI & Knowledge

In this category: tinyhumansai--openhuman (excellent, skip) — general personal AI super-intelligence platform; JerryZLiu--Dayflow is first passive screen-capture work-journal entry.

Standing: Dayflow occupies a distinct niche within Personal AI & Knowledge — where openhuman aims to be a proactive AI assistant, Dayflow is a passive observer that turns raw screen history into structured memory; the two are complementary rather than competing.

Evidence Base

Density: 9/10 — Available: full README (8KB), complete repo metadata (stars, forks, issues, dates, license, archived status), topic tags (12), release history (v1.13.1), creation and last-commit timestamps, Homebrew install reference, Trendshift trending badge. Missing: dependency manifest (Package.swift not surfaced), CI workflow configuration, source file structure, issue content detail.

Notes

The project reached ~6K stars in 8 months with no dependency manifest visible and no CI signals — a pattern common in native macOS apps distributed via DMG/Homebrew where Xcode project management and App Store review pipelines substitute for traditional CI. The v1.13 release number in under a year indicates rapid iteration; the 22 open issues suggests an engaged user base without a backlog crisis. The Markdown export feature is the primary PAI integration surface: a skill that watches ~/Library/Application Support/Dayflow/ for new exports and ingests them into WORK/ would require minimal code and zero changes to Dayflow itself.