factSocial

Fact-check the feed,
on your terms.

A browser extension for Firefox and Chrome that adds a Fact-check button to every Threads and Bluesky post. Powered by an LLM you control, running locally in LM Studio or via your own OpenRouter API key, with optional web search. Returns an inline verdict with sources, confidence, and a per-claim breakdown.

Add to Firefox
Add to Chrome Coming soon
Setup guide How it works GitHub
Firefox is live on Mozilla Add-ons. Chrome users can still load it unpacked until the Chrome Web Store listing publishes.
wellnessdaily ✓ · 2h
BREAKING: Drinking 3 cups of coffee a day adds 10 years to your life, according to a new study.
8.4K 1.2K 412 89 Fact-check
Misleading 68% confidence
Several large studies have linked moderate coffee consumption with lower mortality, but the "adds 10 years" framing dramatically overstates the effect size found in any peer-reviewed research.
▼ Claims (3)
  • Moderate coffee is associated with reduced mortality. MostlyTrue. UK Biobank and EPIC cohort studies show modestly lower all-cause mortality at 3–4 cups/day.
  • 3 cups a day adds 10 years to your life. False. Reported gains in the literature range from a few months to roughly 1.8 years, not a decade.
  • "BREAKING" framing implies new findings. Misleading. The underlying coffee–longevity associations have been studied for decades.
Caveats: Observational studies show correlation, not causation. Effects vary by individual, genetics, and what's added to the coffee.
Sources:
bmj.com harvard.edu fda.gov
via local LLM · factSocial

What you get on every check

One click on the Fact-check button next to a post, and an LLM walks through the claims, hits the web, and returns a structured verdict that renders directly under the post.

Structured verdicts

True, MostlyTrue, Mixed, Misleading, False, Unverifiable, or Opinion. Each comes with a confidence score, summary, per-claim breakdown, and caveats.

Web-aware

The model can call a web_search tool to verify time-sensitive claims. The live card shows every query and its top results as the search runs.

Yours, end to end

No middleman server, no telemetry. Post text goes only to the LLM provider you chose. Search queries go only to the search provider you chose.

Set up in four steps

The whole thing takes about five minutes. You'll install the extension, open its preferences, pick an LLM provider, and (optionally) wire up a search provider.

Install the extension

Firefox is live on Mozilla Add-ons. The Chrome Web Store listing is still pending, so for now:

  • Firefox 115+: click Add to Firefox at the top of the page to install directly from Mozilla Add-ons.
  • Chrome: grab factsocial-chrome-0.2.0.zip from the GitHub repo, unzip it, then visit chrome://extensions, enable Developer mode, and click Load unpacked on the extracted folder.

Either way, the Fact-check button will start appearing on every post on threads.net and bsky.app.

Open factSocial's preferences

All of the configuration lives in your browser's extension settings page. To open it:

  • Firefox: visit about:addons, click factSocial, then click the ⋯ menu and choose Preferences.
  • Chrome: visit chrome://extensions, find factSocial, click Details, then scroll to Extension options.
Firefox about:addons page showing the factSocial extension card with the ⋯ menu open and Preferences highlighted.
Firefox shown above. The Preferences (Firefox) or Extension options (Chrome) entry opens the settings page used in the next two steps.

Pick an LLM provider

factSocial supports two routes. Pick whichever fits your priorities; you can switch any time in the extension's settings.

LM StudioLOCAL

Runs entirely on your machine. No API key, no usage cost, no third-party transmission of post text.

  1. Install LM Studio. In its model search, grab qwen/qwen3.5-9b: a great tool-calling model that runs comfortably on most laptops.
  2. Open the Local Server tab, load qwen3.5-9b, and click Start Server.
  3. Enable the CORS toggle in server settings.
  4. In factSocial settings, leave provider on LM Studio. Click Test connection.

OpenRouterCLOUD

One API key, 200+ models including free options. Start free with minimax/minimax-m2.5:free. No credit needed, and surprisingly good for fact-checking.

  1. Sign up at openrouter.ai/keys and create a key (no payment method required for free models).
  2. In factSocial settings, switch provider to OpenRouter, paste the key.
  3. Set the model to minimax/minimax-m2.5:free to start. Click Test connection.
  4. Want to upgrade later? Add credit and switch to GPT-4o, Claude 3.5, Gemini 2.0, or any of the 200+ paid models. Pennies per check.

Add a web search provider RECOMMENDED

The model can use search to verify time-sensitive claims (current events, breaking news, recent statistics). Without search, it falls back to whatever's in the model's training data, which may be months or years old.

Tavily (recommended)

Tavily ranks results specifically for LLM consumption, so the model gets relevant snippets without scraping. Free tier covers 1,000 searches per month, far more than typical fact-checking needs.

  1. Sign up at tavily.com and grab an API key from the dashboard.
  2. In factSocial settings, scroll to Web search.
  3. Check Enable web search, choose Tavily as the provider, paste the key.
  4. Click Test search to confirm. You'll see the top result echoed back.

Other options: DuckDuckGo (no API key, default; works without setup but is the most fragile and rate-limit prone), or Brave Search (free 2,000 queries/month with a key from brave.com/search/api).

Where your data goes

The extension has no backend. Two destinations only, both chosen by you, both visible in the live progress card on every check.

Post text Sent to your chosen LLM provider on click: LM Studio (localhost) or OpenRouter. Never to factSocial.
Search queries Sent to your chosen search provider (DuckDuckGo, Brave, or Tavily) when the model invokes web_search.
API keys Stored in browser.storage.local and only sent to the provider they belong to. Not transmitted anywhere else.
Telemetry None. The extension has no server, no analytics, no error reporting service.