Inspiration
Most dictionary and translation tools solve the immediate question, but the words and expressions we look up are easily forgotten. Phrase Dictionary AI turns each lookup into a growing personal collection, preserving pronunciation, meaning, usage notes, interpretation candidates, and context for later study.
What it does
Phrase Dictionary AI is a resident Windows application for copied Japanese, English, and romanized Japanese text.
- Copy a word, phrase, or passage in any application.
- Press V twice to open a compact analysis window.
- Review the clipboard text and explicitly choose when to send it for AI analysis.
- Save translations, readings, pronunciation guides, explanations, interpretation candidates, and tags to a local personal dictionary.
- Search and revisit the collection as it grows.
For longer passages, the app identifies registered vocabulary, analyzes unregistered words in batches, stores full-text translations separately, and exports a self-contained offline HTML study document. Readers can switch between word-by-word and full translation views, jump from a word to its dictionary card, and return to the original reading position.
Why it is different
The goal is not merely to keep translation history. It creates a collection of expressions the user has personally encountered. Seeing known words highlighted and watching the dictionary grow makes vocabulary building feel more like collecting than administrative study work.
Privacy and safety
- Clipboard text is not sent until the user presses the analysis button.
- The Gemini API key is stored in Windows Credential Manager, not SQLite or logs.
- Dictionary data stays in a local SQLite database.
- Exported HTML and JSON exclude API keys, internal prompts, logs, and settings.
- The public repository and release package exclude local databases, logs, credentials, and virtual environments.
How Codex with GPT-5.6 was used
Codex with GPT-5.6 was the main development partner during roughly five hours of focused development. It accelerated requirements analysis, architecture decisions, Python and web UI implementation, debugging, regression testing, documentation, security review, release packaging, and GitHub publication.
Important decisions made during this workflow included keeping runtime data local, storing credentials in Windows Credential Manager, separating full-text translations from dictionary entries, making cached results language-aware, labeling previously saved contextual explanations clearly, and producing a reproducible screenshot-based demo-video pipeline. Codex also helped trace UI navigation and cache bugs, add targeted regression tests, scan the release for credentials, and prepare the public GPL-3.0-only release.
Codex with GPT-5.6 was used to build the project; it is not the runtime translation model. The distributed application uses the user's own Google Gemini API key for language analysis.
Technology
PySide6 and Qt WebEngine provide the Windows UI, FastAPI/Uvicorn provide the local API, SQLite stores the dictionary, Janome handles Japanese tokenization, and Google Gemini performs runtime language analysis.
Availability
Setup and testing instructions, source code, and a downloadable Windows release are available in the public GitHub repository. Users supply and manage their own Gemini API key. AI-generated translations and explanations may require user review.
Log in or sign up for Devpost to join the conversation.