FluentRead bilingual reading and translation setup

FluentRead is an open-source browser extension that displays translations beneath the original text and provides selection translation, AI reading cards, learning records, document reading and bilingual captions on supported platforms. Its inputs are webpage text, selected sentences, documents or readable captions. Available output and controls depend on the installed build, browser and selected translation service.

Basic webpage translation can use free services without an API key. AI explanations need an available AI service and model. A local model changes the processing path for its assigned feature; it does not establish that every tool runs offline.

The official README supplies the installation links below. After installing for the relevant browser, refresh pages that were already open and pin the FluentRead toolbar icon.

In Chrome, find FluentRead in the extensions menu beside the address bar and select its pin icon. Other browsers may place this control elsewhere; first confirm that the FluentRead menu opens.

Opening a selection card does not automatically call AI. A deliberate learning action sends the selection and permitted context to the configured service.

Choose a service under Settings → Translation services and enter its URL, key and required provider fields; an AI service also needs a model available to the account. Check connection sends a short request and may use allowance. Success describes that check only.

1. Confirm language support, API allowance, region and model access with the provider.

2. Configure and check the connection, then choose Set as default translation service.

3. Test a short ordinary passage before translating a full article.

4. Documents, subtitles and reading cards can use separate services, so check each active selection.

Automatic balancing or fallback in the free service may send the same passage to another provider. To restrict text to a single provider, disable other candidates or select that provider's standalone service.

Extra AI context and AI multi-paragraph translation are off by default. Context submits more webpage text and may increase allowance use and waiting time. Its range is limited, so verify terminology and important information.

Ollama's default address is http://127.0.0.1:11434 without an API key. Other devices or ports need a complete /v1/chat/completions URL as documented. On slower computers, set concurrency to 1 under Request limits → Entire service → Custom, then observe response times and resource use.

The first local-model download still needs a connection, and download size is not runtime memory use. Model files remain in the current browser and are excluded from configuration backups. Ollama manages its own model residency.

Document import and download

Open Document translation and import a file or provide a direct online PDF link. If a login prevents the link from opening, download the file and import it locally.

Documents can pause, continue and retry while retaining completed results and edits. After changing language or some settings, Retranslate replaces existing translations and manual edits. Download the current version first if it needs to be kept.

Untranslated passages keep their original text in partial downloads, so a download does not establish full translation. Check recognition, formulas and specialist terms against the source file. Clearing browser data can also remove local documents and edits.

Video and meeting captions

The official guide lists YouTube, X, Udemy, Disney+ and Google Meet, Teams and Zoom web meetings. Most routes need readable captions supplied by the platform. Account, region, player and meeting-host permissions can restrict availability.

Enable Video subtitle translation, then enable the player's native captions. For X videos without captions, supported desktop Chrome, Edge and Firefox extensions can use Whisper in the browser. Models download on first use, and the documented limit is 20 minutes per video.

Whisper recognition can run locally, but recognized text may still go to the selected translation service. FluentRead does not create captions for meetings that supply none and does not control desktop meeting applications.

Userscripts require a compatible manager such as Tampermonkey, Violentmonkey, Via or Safari Userscripts. Keep a single FluentRead script enabled, check site access and the script switch, then reload an ordinary webpage.

The native Safari preview still lacks Chrome's built-in translation, image OCR, area capture and some platform features. iOS system video fullscreen does not display normal webpage overlays. The system-caption route described in the guide still needs physical-device verification.

Sentence roles and parts of speech describe different things, and model formatting can cause analysis to fall back to text. Treat the result as a reading aid and compare it with the sentence and context. Use a glossary when important terms need consistent translation.

Use an ordinary public article with Traditional Chinese as the target language. Start with free translation without an API key. AI explanations remain optional for users who already have an available service.

This procedure follows the official interface documentation; this research did not run the extension. When testing it, record the browser, installed version, service, language and article URL, with private content removed from screenshots.

Settings and learning records stay local by default. Cloud translation sends relevant text to the chosen provider, and fallback may send it to another provider. Extra AI context, documents and subtitles have separate processing paths, so check the feature and service before using sensitive material.

API keys persist in local settings instead of disappearing when the browser closes. Configuration storage uses public application material for encryption, so it does not provide secrecy under a user-owned password. People with access to the browser profile or backups may still read credentials.

Google Drive and WebDAV configuration backups are manual; changing settings does not upload them automatically. Consent for sensitive settings stays in the current browser until that option is disabled. Cancellation, failure or changing accounts does not clear the choice automatically.

The official backup guide and privacy guide explain that disabling sensitive settings, revoking access or uninstalling does not delete old cloud files. To remove existing secrets, confirm a basic-settings save or merge and manage cloud version history and other copies separately.

Free endpoints, models and browser restrictions can change, and a documented feature may not have reached the installed store build. Report the installed version and publicly reproducible steps. Verify important recognition, translation and AI-explanation output manually.

Readers can use a store build. To investigate or change the code, use the checked commit, Node.js 22.23.4 from .nvmrc and the package's pnpm 12.18.0 through Corepack.

corepack pnpm dev starts local development, and corepack pnpm zip packages Chrome. These are upstream project commands. This research did not install dependencies, build the extension or load it in a browser.

The maintainer release guide covers Edge, Firefox, Userscript and Safari builds separately. Producing a ZIP does not complete store review or signing. Platforms such as Firefox and Safari have their own distribution requirements.

Complete basic translation and restore before configuring AI, document or subtitle services. To reduce external text transfer, consider a local translation route and check the network behavior of other tools separately.

• Official user guide: check current controls and support.

• Repository Issues: report versions, a public URL, reproduction steps and redacted screenshots.

• GPL-3.0 license: read it with third-party notices before distributing or modifying the program.

The official README welcomes code, documentation, interface translations and website-adaptation contributions. Describe a single reproducible problem with expected and actual results before proposing a matching change.