Daily is a local-first, day-centric task manager for macOS with a built-in AI assistant that executes real operations instead of just answering questions.
Tasks are organized by date rather than in endless lists, live on a three-column Kanban board, and are written in full Markdown. All data is stored in SQLite on your Mac; iCloud sync is optional. The AI assistant runs an agent loop over the same task operations the UI uses — with local models that run entirely in-app with zero extra setup, or optionally any OpenAI-compatible remote API.
Install with Homebrew:
brew install --cask scheron/tap/dailyOr install manually:
- Open Releases
- Download the macOS
.dmg(Apple Silicon) - Move Daily.app to Applications
The app checks GitHub Releases and updates itself in place, so you only download once.
Builds are ad-hoc signed, not notarized. If macOS blocks the first launch, run:
xattr -rd com.apple.quarantine /Applications/Daily.app- planning around a single question: what did I do yesterday, and what matters today
- moving tasks through a board (active / discarded / done) instead of maintaining lists
- writing task content in real Markdown with code, tables, and attachments
- keeping separate task spaces per project and switching between them instantly
- tracking where the time goes — estimates, time spent, and activity stats per day
- delegating routine task operations to an AI assistant that acts, not chats
- owning the data: plain SQLite on disk, no account, no mandatory cloud
Every task belongs to a date. The calendar is the primary navigation surface and shows at a glance which days carry open work; tasks reschedule via drag-and-drop onto a date.
SQLite on your Mac is the single source of truth. The app is fully functional offline. Sync, when enabled, treats iCloud as a transport — never as the primary store.
The assistant executes the same operations the UI exposes: creating and editing tasks, changing statuses, managing tags and time, searching by date, summarizing the day. Destructive operations always suspend and ask for explicit confirmation before running.
- three-column Kanban board (active / discarded / done) with drag-and-drop reordering and cross-status moves
- a resizable slide-in editor with property rows: status, date, project, estimate, time spent, tags
- prev/next navigation between tasks without leaving the editor
- quick task actions: duplicate, copy ID or content, delete
- unsaved changes are guarded with a save/discard prompt
The editor is CodeMirror-based with live-preview behavior:
- syntax highlighting for fenced code blocks, with a copy button and auto-closing fences
- Markdown tables render as live, click-to-edit tables
- slash commands for inserting headings, lists, quotes, checkboxes, and other blocks
- wrapped list items keep a hanging indent; ordered lists renumber automatically
- file and image attachments, stored as regular files on disk
- inline tag commands (
#tag,-#tag) with autocomplete
Projects (branches) isolate task spaces, with instant switching, inline project creation, and task moves between projects. Search resolves matches across all projects and switches context automatically.
Every meaningful task change — created, completed, discarded, reactivated, edited, moved, restored, deleted — is recorded in an append-only event log that syncs across devices.
- the Activity widget shows a day-by-day feed of what changed
- same-day restore for deleted tasks, directly from the feed
- a per-task timeline of every change
Soft-deleted items past the retention window are cleaned up automatically on app start.
Fuzzy full-text search across all tasks: tolerates typos, matches on content, and jumps directly to the task's date (switching project if needed).
Per-day and per-period statistics are computed in SQL and rendered in the sidebar widget: a resolution ring, a tag-composition ring, weekday and hour histograms, and peak hour / peak weekday / top tag for the period, with a week/month toggle.
- light, dark, or follow-the-system mode with 9 accent color presets
- configurable sidebar widgets (calendar, stats, activity) with drag-to-reorder
Sync is off by default and optional. When enabled, Daily syncs snapshots through iCloud Drive with last-write-wins merge and a periodic auto-sync. iCloud placeholder files are recognized and downloaded before being treated as data. The app never requires an account or a server.
The assistant is not a chat bolted onto the app — it is a full agent with tool calling. It runs in its own window and operates on the workspace through the same storage layer as the UI: task CRUD, status updates, date-based search, batch actions, tag/time/attachment management, and day summaries.
How it works:
- each message runs an agent loop: the model calls tools, observes results, and iterates until done
- responses stream live, with model reasoning shown in a collapsible timed panel
- destructive tool calls suspend until explicitly confirmed in the UI
- conversations persist across app restarts; long histories are compacted automatically
- a live indicator tracks how full the model's context window is
- the assistant can read web pages on request: private/internal addresses are blocked, fetched content is treated as untrusted, and the first fetch asks for permission
Typical prompts:
- "Create a task to review the PR tomorrow at 2pm"
- "I spent 2 hours on the documentation task"
- "Complete all my work tasks for today"
- "How's my day going?"
Both local and remote models are supported, switchable from the same settings screen — and both run the full agent loop.
Local — zero-setup and fully offline. No Ollama, LM Studio, or command line required: models are downloaded from a curated in-app catalog and are ready to use immediately — Daily manages its own llama.cpp server with Metal acceleration under the hood. The catalog includes the Qwen3.5 family, GLM-4 9B, Mistral Nemo 12B, and Llama 3.1 8B. Downloads are resumable and sha256-verified; idle models unload automatically to free memory (configurable, 15 minutes by default).
Remote — any OpenAI-compatible API with your key. Presets for OpenAI and DeepSeek models, or a custom base URL. Models without native function calling (Qwen 3.5-style fine-tunes) are supported through a compatibility mode.
- all data is stored locally in SQLite
- attachments are regular files on disk, tracked by the database
- no account, no telemetry, no mandatory cloud dependency
- deletes are soft: records are tombstoned, recoverable same-day, and purged after a retention window
- with sync enabled, data leaves the machine only as snapshots in your own iCloud Drive
macOS (Apple Silicon) only. Electron would allow Windows/Linux builds, but they are not shipped due to limited testing coverage (or I'm just lazy 🤷♂️).