toXic Filter

It classifies posts on X with Jev and hides the expressions you chose to avoid behind a translucent overlay. Categories include discrimination, provocation, hatred or contempt, calls to attack, cynicism and resignation; each hidden post can be revealed again with the ✗ button.

Download the latest ZIP from releases and unzip it, open chrome://extensions in Chrome 120 or later, enable developer mode, and load the unpacked folder containing manifest.json. Then open the extension's options, save a TypeSafe API token, open X and click the icon to start.

Yes. The README says you need your own Jev API token, created at console.typesafe.ai/keys, and that API usage is billed to your TypeSafe account. The token is stored in the extension's storage.local and is not exposed to the content script or Chrome Sync.

Post text near the viewport and quoted text are sent directly to TypeSafe. That can include names and URLs inside the text, and posts from visible non-public accounts are also in scope. Images, videos and DMs are not. The README advises pausing the extension before viewing content you do not want sent.

It sends jev-latest seven questions: the six categories plus an insufficient-context check. A post is hidden when an enabled category's probability passes the threshold — 70% by default — unless insufficient context is at or above 65%; posts whose text and quote together exceed 6,000 characters are not classified.

At most 300 submissions per minute across all tabs, 8 in parallel, with one retry on 429 or 529 and new submissions stopped after API errors. Results stay in memory for up to 10 minutes or 1,000 items, and up to 500 per tab. After reloading X, click the icon to resume.

Yes, with Node.js 22.13 or later: npm ci, npm run build to generate dist/extension, and npm run check to test without calling the API and produce a ZIP with a SHA-256 checksum. Creating the ZIP needs zip and unzip on macOS or Linux; npm run package also emits dist/toxic-filter.zip.