Dayflow

Introduction

Keeping a work journal is useful for recalling what you accomplished, but doing it manually is tedious and easy to forget. Dayflow automates this by running quietly in the background on macOS, capturing periodic snapshots of your screen, and using AI vision models to generate a timeline of your activities throughout the day.

What Dayflow Does

Dayflow takes screenshots at configurable intervals and feeds them to a vision-capable language model to produce human-readable summaries of your activity. These summaries are organized into a chronological timeline you can review at any point. All data stays on your machine by default when using local AI backends, preserving privacy.

Architecture Overview

Dayflow is a native macOS application. It uses the macOS screenshot APIs to capture the screen periodically. Captured images are processed through configurable AI backends -- either local models running via Ollama or LM Studio, or cloud-hosted LLMs via API. The results are stored in a local database and presented through a timeline interface. The architecture is designed so that no data leaves your machine unless you explicitly configure a cloud AI provider.

Self-Hosting & Configuration

Dayflow runs entirely on your local Mac. After installation, configure your preferred AI backend in settings. For fully private operation, set up Ollama or LM Studio with a vision-capable model. You can also configure cloud LLM providers by entering an API key. Screenshot intervals, storage location, and which apps or screens to capture can all be adjusted in the preferences.

Key Features

• Automatic periodic screenshot capture in the background
• AI-powered activity summarization using vision models
• Timeline view of daily work accomplishments
• Local-first design with all data stored on your machine
• Support for local AI via Ollama and LM Studio
• Optional cloud LLM integration for summarization
• Configurable capture intervals and app filtering
• Privacy-focused architecture with no mandatory cloud dependency
• Native macOS menu bar application
• Searchable activity history

Comparison with Similar Tools

Compared to Rewind (now Limitless), Dayflow is fully open source and can run with entirely local AI, avoiding any cloud data transmission. Unlike manual time-tracking tools such as Toggl or Clockify, Dayflow is automatic and requires no user input. Compared to ActivityWatch, which tracks app usage time, Dayflow uses AI vision to generate descriptive summaries of what you were actually doing, not just which apps were open.

FAQ

Is my screen data sent to the cloud? Only if you configure a cloud LLM provider. When using Ollama or LM Studio as the AI backend, all processing happens locally and no screenshots leave your machine.

What AI models work with Dayflow? Any vision-capable model compatible with Ollama or LM Studio works for local processing. For cloud usage, models with vision support from providers like OpenAI or Anthropic can be configured.

Does it impact system performance? Dayflow is designed to be lightweight. Screenshot capture is brief and infrequent, and AI processing can be scheduled during idle periods to minimize impact.

Sources

• GitHub repository: https://github.com/JerryZLiu/Dayflow