fthr is a customizable word processor that brings your documents, AI agents, and reusable macros into one workspace. It runs in your browser, giving you a place to write, organize, and revise your work.
Send selected passages directly to AI, build prompts that draw on your notes and document metadata, and configure agents with your own instructions and model settings. Save macros for the writing and editing tasks you repeat, from developing ideas to revising a draft.
Keep your documents local in your browser, and connect your own API key for AI assistance. AI requests send the relevant text and context to your configured provider.
Search is a central part of fthr's design: the header Search is the single place to find your writing, chats, models, settings, and actions. The library has no separate search. All settings options are being folded into the Search menu, including page setup and document export, to keep the writing workspace clear and make controls easy to find by name.
Open Search or press Ctrl/Cmd+K to search directly or browse Settings. Settings should be discoverable through this shared menu as the application grows.
| Feature | What you can do |
|---|---|
| Instant macros and selection sending | Run saved prompts against a selected passage, with individual model settings for each macro. |
| Saved agent profiles | Name writing assistants, save their instructions, model settings, tools, and approval preferences, and choose a default. Start fresh chats from any profile. |
| Context-aware prompts | Include selected text, named notes, and document metadata in your prompts, using references that pull in the relevant content. |
| Writing versions and notes | Create, copy, rename, and switch writing versions from the right-pane Versions section or Search. Keep shared notes beside your writing and organize documents into folders. |
| Workspace search | Open Search or press Ctrl/Cmd+K to find document text, writing versions, notes, chat titles and messages, profiles, settings, and models in one command palette. Search runs locally. |
| Custom document metadata | Keep tags, descriptions, and custom fields with a document, and reference those fields in AI prompts. |
| Semantic and hybrid search | Search by keyword, by meaning through embeddings, or combine both approaches to retrieve relevant passages. Embedding search requires configuration. |
| Checkpoints and version diffs | Save named checkpoints in the History sidebar, preview saved text, and compare it with your current writing. |
Open fthr.vercel.app to start writing. AI assistance is optional; the editor works without an API key.
To use AI, open Search → Settings → Manage API keys, add a named OpenAI connection, and load its chat models. New workspaces start with an empty model catalog. Open Search → Agent profiles to create a named assistant, choose its model and settings, and set a default. Profiles are searchable by name and appear at the top of New chat (Ctrl+M), followed by available models. The Chats tab lists conversations. Each chat keeps a copy of its profile settings; profile changes affect future chats unless you explicitly apply an update. See Agent profiles. API usage is billed by your configured provider.
Documents and preferences are stored in your browser, primarily through IndexedDB. API keys and cached model catalogs use separate device storage and are excluded from workspace exports and Google Drive sync. This browser storage is accessible to the application and is not an encrypted vault. Data belongs to that site's browser profile; opening another browser, device, or local installation does not automatically bring it across. See API connections and catalogs.
AI features send relevant text and context to the configured API provider. Embedding requests also send text when enabled. Local storage does not mean AI processing stays on your device.
Use Search → Settings → Import / Export workspace to download a JSON export regularly, especially before clearing site data or changing browsers. The current export includes documents, folders, prompts, the default chat configuration, and agent profiles with the default profile choice. Imported profiles are added as new copies; an existing default is preserved. API credentials remain on their original device, so restored profiles may need to be reconnected. Separate agent conversations and checkpoint history are not included, so it is not a complete workspace backup.
Use Node.js 24.15.0 or newer within 24.x and npm 11.19.0. The repository checks supported versions during installation.
git clone https://github.com/darrylschaefer/ai-text-editor.git
cd ai-text-editor
npm ci
npm run devOpen localhost:5173. An API key is needed for AI features, not for local editing or the automated test suite.
No environment file is required for the default setup. Optional configuration is listed in .env.example. VITE_* values are included in the browser build; use the in-app API setting for a personal key rather than embedding it in a public deployment.
| Command | Purpose |
|---|---|
npm run dev |
Start the development server. |
npm run check |
Type-check the application and tests, run the test suite, and build the web app. |
npm run test:e2e |
Build and run desktop and phone browser chat journeys. Install browsers first with npx playwright install chromium firefox webkit. |
npm run check:all |
Run all type checks, unit/integration tests, the production build, and browser tests. |
npm run test:watch |
Run tests while making changes. |
npm run build |
Create the production web build in dist/. |
npm run preview |
Serve the production build locally after building it. |
Pull requests and updates to main run validation on Linux and Windows. The Linux validation job also builds the web Docker image; a separate Linux browser job runs the chat journeys across desktop and phone projects. See browser testing for setup and diagnostics, and CONTRIBUTING.md for the branch/PR workflow, individual checks, Docker, and desktop development notes.
fthr is actively developed. Persistence recovery, complete backups, and AI editing safeguards remain areas of ongoing work. Native desktop packaging is not covered by the web validation checks.
Found a problem or have an idea? Open an issue. The engineering backlog records known follow-ups.
Built by Darryl Schaefer. Released under the MIT License.
Inspired by and adapted from BetterChatGPT, whose original work is released under CC0.