Quickstart • Why I built Dayflow • Features • How it works • Installation • Data & Privacy • Automation • Debug & Developer Tools • Auto‑updates • Contributing

Dayflow is a native macOS app (SwiftUI) that records your screen at 1 FPS, analyzes it every 15 minutes with AI, and generates a timeline of your activities with summaries. It's lightweight (25MB app size) and uses ~100MB of RAM and <1% cpu.

Privacy‑minded by design: You choose your AI provider. Use Gemini (bring your own API key), local models (Ollama / LM Studio), or ChatGPT/Claude (requires paid subscription). See Data & Privacy for details.

I built Dayflow after realizing that my calendar wasn't the source of truth for how I actually spent my time. My screen was. I wanted a calm, trustworthy timeline that let me see my workday without turning into yet another dashboard I had to maintain.

Dayflow stands for ownership and privacy by default. You control the data, you choose the AI provider, and you can keep everything local if that's what makes you comfortable. It's MIT licensed and fully open source because anything that watches your screen all day should be completely transparent about what it does with that information. The app should feel like a quiet assistant: respectful of your attention, honest about what it captures, and easy to shut off.

• Automatic timeline of your day with concise summaries.
• 1 FPS recording - minimal CPU/storage impact.
• 15-minute analysis intervals for timely updates.
• Watch timelapses of your day.
• Auto storage cleanup - configurable storage limits.
• Distraction highlights to see what pulled you off‑task.
• Timeline export — export your timeline as Markdown for any date range.
• Native UX built with SwiftUI.
• Auto‑updates with Sparkle (daily check + background download).

Set intentions, reflect on your day, and get AI-generated summaries of your activity.

• Morning intentions — plan what you want to accomplish.
• Evening reflections — review how your day actually went.
• AI summaries — get auto-generated insights from your timeline.
• Scheduled reminders — configurable notifications for intentions and reflections.
• Weekly view — see patterns across your week.

Note: Journal is currently in beta with limited access. Enter your access code in the app to unlock it.

• Infinitely customizable dashboard — ask any question about your workday, pipe the answers into tiles you arrange yourself, and track trends over time.

  

1. Capture — Records screen at 1 FPS in 15-second chunks.
2. Analyze — Every 15 minutes, sends recent footage to AI.
3. Generate — AI creates timeline cards with activity summaries.
4. Display — Shows your day as a visual timeline.
5. Cleanup — Auto-manages storage based on your configured limits (1GB–20GB or unlimited).

The efficiency of your timeline generation depends on your chosen AI provider:

flowchart LR subgraph Gemini["Gemini Flow: 2 LLM Calls"] direction LR GV[Video] --> GU[Upload + Transcribe<br/>1 LLM call] --> GC[Generate Cards<br/>1 LLM call] --> GD[Done] end subgraph Local["Local Flow: 33+ LLM Calls"] direction LR LV[Video] --> LE[Extract 30 frames] --> LD[30 descriptions<br/>30 LLM calls] --> LM[Merge<br/>1 call] --> LT[Title<br/>1 call] --> LC[Merge Check<br/>1 call] --> LMC[Merge Cards<br/>1 call] --> LD2[Done] end subgraph ChatCLI["ChatGPT/Claude Flow: 4-6 LLM Calls"] direction LR CV[Video] --> CE[Extract frames<br/>every 60s] --> CB[Batch describe<br/>10 frames/call] --> CM[Merge segments<br/>1 call] --> CC[Generate Cards<br/>1 call] --> CD[Done] end %% Styling classDef geminiFlow fill:#e8f5e8,stroke:#4caf50,stroke-width:2px classDef localFlow fill:#fff8e1,stroke:#ff9800,stroke-width:2px classDef chatcliFlow fill:#e3f2fd,stroke:#1976d2,stroke-width:2px classDef geminiStep fill:#4caf50,color:#fff classDef localStep fill:#ff9800,color:#fff classDef chatcliStep fill:#1976d2,color:#fff classDef processing fill:#f5f5f5,stroke:#666 classDef result fill:#e3f2fd,stroke:#1976d2 class Gemini geminiFlow class Local localFlow class ChatCLI chatcliFlow class GU,GC geminiStep class LD,LM,LT,LC,LMC localStep class CB,CM,CC chatcliStep class GV,LV,LE,CV,CE processing class GD,LD2,CD result 

Gemini leverages native video understanding for direct analysis. Local models reconstruct understanding from individual frame descriptions. ChatGPT/Claude uses CLI tools to batch-process extracted frames with frontier reasoning models—balancing quality and efficiency.

Download (end users)

1. Grab the latest Dayflow.dmg from GitHub Releases.
2. Open the app; grant Screen & System Audio Recording when prompted:
   macOS → System Settings → Privacy & Security → Screen & System Audio Recording → enable Dayflow.

Build from source (developers)

1. Install Xcode 15+ and open Dayflow.xcodeproj.
2. Run the Dayflow scheme on macOS 13+.
3. In your Run scheme, add your GEMINI_API_KEY under Arguments > Environment Variables (if using Gemini).

• macOS 13.0+
• Xcode 15+
• A Gemini API key (if using Gemini): https://ai.google.dev/gemini-api/docs/api-key

1. Download Dayflow.dmg and drag Dayflow into Applications.
2. Launch and grant the Screen & System Audio Recording permission.

git clone https://github.com/JerryZLiu/Dayflow.git cd Dayflow open Dayflow.xcodeproj # In Xcode: select the Dayflow target, configure signing if needed, then Run.

If you are using Homebrew, you can install Dayflow with:

$ brew install --cask dayflow

This section explains what Dayflow stores locally, what leaves your machine, and how provider choices affect privacy.

All Dayflow data is stored in: ~/Library/Application Support/Dayflow/

• Recordings (video chunks): Dayflow/recordings/ (or choose "Open Recordings..." from the Dayflow Taskbar Icon Menu)
• Local database: Dayflow/chunks.sqlite
• Recording details: 1 FPS capture, analyzed every 15 minutes, configurable storage limits
• Purge / reset tip: Quit Dayflow. Then delete the entire Dayflow/ folder to remove recordings and analysis artifacts. Relaunch to start fresh.

• Gemini (cloud, BYO key) — Dayflow sends batch payloads to Google's Gemini API for analysis.
• Local models (Ollama / LM Studio) — Processing stays on‑device; Dayflow talks to a local server you run.
• ChatGPT / Claude (CLI-based, paid plan required) — Dayflow drives the Codex CLI (ChatGPT) or Claude Code CLI directly on your Mac. Requires an active ChatGPT Plus/Pro or Claude Pro subscription. Uses frontier reasoning models for best-in-class narrative quality.

• Short answer: There is a way to prevent Google from training on your data. If you enable Cloud Billing on at least one Gemini API project, Google treats all of your Gemini API and Google AI Studio usage under the “Paid Services” data‑use rules — even when you’re using unpaid/free quota. Under Paid Services, Google does not use your prompts/responses to improve Google products/models.
  • Terms: “When you activate a Cloud Billing account, all use of Gemini API and Google AI Studio is a ‘Paid Service’ with respect to how Google Uses Your Data, even when using Services that are offered free of charge.” (Gemini API Additional Terms)
  • Abuse monitoring: even under Paid Services, Google logs prompts/responses for a limited period for policy enforcement and legal compliance. (Same Terms)
  • EEA/UK/Switzerland: the Paid‑style data handling applies by default to all Services (including AI Studio and unpaid quota) even without billing. (Same Terms)

A couple useful nuances (from docs + forum clarifications):

• AI Studio is still free to use; enabling billing changes data handling, not whether Studio charges you. (Pricing page)
• UI “Plan: Paid” check: In AI Studio → API keys, you’ll typically see “Plan: Paid” once billing is enabled on any linked project (UI may evolve).
• Free workaround: “Make one project paid, keep using a free key elsewhere to get the best of both worlds.” The Terms imply account‑level coverage once any billing account is activated, but the Apps nuance above may limit this in specific UI contexts. Treat this as an interpretation, not legal advice.

• Privacy: With Ollama/LM Studio, prompts and model inference run on your machine. LM Studio documents full offline operation once models are downloaded.
• Quality/latency: Local open models are improving but can underperform cloud models on complex summarization.
• Power/battery: Local inference is GPU‑heavy on Apple Silicon and will drain battery faster; prefer plugged‑in sessions for long captures.
• Future: We may explore fine‑tuning or distilling a local model for better timeline summaries.

References:

• LM Studio offline: https://lmstudio.ai/docs/app/offline
• Ollama GPU acceleration (Metal on Apple): https://github.com/ollama/ollama/blob/main/docs/gpu.md

• Privacy: Your screen data is processed by OpenAI (ChatGPT) or Anthropic (Claude) depending on which CLI you configure. Review their respective privacy policies.
• Quality: Frontier reasoning models provide the highest quality narratives and summaries.
• Subscription required: You must have an active paid subscription (ChatGPT Plus/Pro at $20+/month, or Claude Pro at $20/month). The CLI tools authenticate through your existing subscription.
• Setup: Requires installing the Codex CLI or Claude Code and staying signed in.
• Internet: Requires an active internet connection (no offline mode).

To record your screen, Dayflow requires the Screen & System Audio Recording permission. Review or change later at:
System Settings → Privacy & Security → Screen & System Audio Recording.
Apple’s docs: https://support.apple.com/guide/mac-help/control-access-screen-system-audio-recording-mchld6aa7d23/mac

• AI Provider
  • Choose Gemini (set API key), Local (Ollama/LM Studio), or ChatGPT/Claude (install CLI + paid subscription).
  • For Gemini keys: https://ai.google.dev/gemini-api/docs/api-key
  • For ChatGPT: Install Codex CLI, sign in with your ChatGPT Plus/Pro account
  • For Claude: Install Claude Code, sign in with your Claude Pro account
• Capture settings
  • Start/stop capture from the main UI. Use Debug to verify batch contents.
• Data locations
  • See Data & Privacy for exact paths and a purge tip.

Dayflow registers a dayflow:// URL scheme so you can trigger common actions from Shortcuts, hotkey launchers, or scripts.

Supported URLs

• dayflow://start-recording — enable capture (no-op if already recording)
• dayflow://stop-recording — pause capture (no-op if already paused)

Quick checks

• From Terminal: open dayflow://start-recording or open dayflow://stop-recording
• In Shortcuts: add an Open URLs action with either link above

Deeplink-triggered state changes are logged as reason: "deeplink" in analytics so you can distinguish automations from manual toggles.

You can click the Dayflow icon in the menu bar and view the saved recordings

Dayflow integrates Sparkle via Swift Package Manager and shows the current version + a “Check for updates” action. By default, the updater auto‑checks daily and auto‑downloads updates.

Dayflow/ ├─ Dayflow/ # SwiftUI app sources (timeline UI, debug UI, capture & analysis pipeline) ├─ docs/ # Appcast and documentation assets (screenshots, videos) ├─ scripts/ # Release automation (DMG, notarization, appcast, Sparkle signing, one-button release) 

• Screen capture is blank or fails
  Check System Settings → Privacy & Security → Screen & System Audio Recording and ensure Dayflow is enabled.
• API errors
  Go into settings and verify your GEMINI_API_KEY and network connectivity.

• V1 of the Dashboard (track answers to custom questions)
• V1 of the daily journal — now in beta!
• Fine-tuning a small VLM for improved local model quality

PRs welcome! If you plan a larger change, please open an issue first to discuss scope and approach.

Licensed under the MIT License. See LICENSE for the full text. Software is provided “AS IS”, without warranty of any kind.

• Sparkle for battle‑tested macOS updates.
• Google AI Gemini API for analysis.
• Ollama and LM Studio for local model support.
• OpenAI Codex CLI and Claude Code for CLI-based inference.