Changelog

Daisy can now start automatically when you log in to your Mac. One command sets it up; she'll run quietly in the background, restart herself if she crashes, and stay out of your way. The setup script is cleaner, there's a proper changelog file (this one), and the API has a "are you alive?" check so other tools can watch Daisy.

You can now say "undo" and Daisy actually undoes the last thing she did — deletes the note, removes the task, cancels the reminder, or forgets the fact you just told her. You can also see a log of everything she's done recently, and ask her current settings (which voice, which AI provider, etc).

Daisy now remembers things across conversations. Tell her once "I prefer Python", "my dog's name is Rex", or "I work at Acme Corp", and she'll still know next week. She also automatically summarizes long conversations so she doesn't lose track of context (or rack up huge token bills) when you've been chatting for hours.

Daisy can now read and write to your Mac's Calendar, Reminders, and Mail apps. Ask her what's on your schedule today, have her book a dentist appointment, list your to-dos, add new reminders, check recent emails, search your inbox, or draft/send an email — all by voice.

Your API keys (OpenAI, Anthropic, etc.) no longer sit in plain-text config files. They go in the macOS Keychain — the same secure storage your Mac uses for WiFi passwords. You manage them from the UI or via API; deleting a key wipes it from the Keychain.

Daisy now feels like a real Mac app instead of a script. There's a small icon in your menu bar that shows whether she's listening, speaking, or idle. Her replies type out word-by-word like a chat app instead of arriving in one chunk. And when she needs microphone or screen-recording permission, she walks you straight to the right pane in System Settings instead of leaving you to find it.

Daisy gets a real web UI (no command-line needed), can listen with better quality voice models that run locally on your Mac, and can be used as a voice front-end by other AI tools (like Claude Code). She also stops hallucinating words from background noise.

Daisy becomes a real voice assistant instead of a "press button, then talk" tool. She listens continuously for her name, lets you cut her off when she's talking too much, actually asks before doing risky things, and fires reminders on time. She can also drive other apps on your desktop (open Xcode, click buttons, take screenshots) and talk to Anthropic's Claude models, not just OpenAI.

A maintenance release — Daisy got noisier and more reliable. Voice replies stopped being silent on launch, her TTS started using the right voice from your config, the safety filter stopped incorrectly blocking harmless commands, and a dozen other rough edges got sanded down. No new features, just everything working as intended.

The first version of Daisy with a proper structure. Instead of one giant Python file, she has clean separate modules for listening, thinking, and doing things. She remembers conversations across sessions in a small database, logs everything she does, and has guardrails against running anything destructive.

The first working version of Daisy — voice-first, monolithic, but actually useful. She had wake-word listening, could control other apps on your desktop, and even had a sub-tool called "Praiser" for tracking compliments. Rough around the edges, but the proof that the idea worked.