𝕏 Sentiment Tracker

This app builds a sentiment tracker with the Grok API that scores what people on 𝕏 say about any topic, post by post and day by day. It's a starting point for brand monitoring or tracking the reaction to a launch.

Give it a topic, like a company, a product, or an event, and it searches 𝕏 for popular posts about it from the last 10 days, filters out the noise, and scores the sentiment. It runs as a small web app or in your terminal, and shows the searches, the posts it keeps, and Grok's reasoning as they happen. It follows the same approach as the 𝕏 sentiment notebook.

• Search 𝕏 with the built-in X Search tool and 𝕏's search operators
• Run one search per day in parallel to collect more posts than a single search returns
• Use citations to keep only the posts that X Search actually returned
• Show server-side searches and reasoning as they happen with the SDK's stream events
• Score every post as structured output, and clamp the scores, since a schema's minimum and maximum aren't enforced
• Use a low reasoning effort to keep each step fast, even when Grok scores every post
• Stream progress from a Node server to a web page with server-sent events, and cancel the work when the page closes

You need Node.js 22.13 or later. Put your API key in .env at the root of the repo, or export XAI_API_KEY.

cd examples/x-sentiment-tracker/typescript
npm install
npm run web

Open http://localhost:3000, enter a topic or pick one, and click Analyze. The page works like a social listening dashboard. It shows each day's search as it runs, how many posts were found and kept, and Grok's reasoning while it scores. Then it shows the overall score from -1 (negative) to +1 (positive), the split between positive, neutral, and negative posts, the sentiment day by day, why Grok gave that score, and every post with its own score. Click Stop or close the page to cancel.

To check a topic from the terminal instead, pass it as an argument. It defaults to SpaceX.

npm start -- SpaceX

The topic goes into the search query as is, so you can use 𝕏's search operators in it, like Starship OR Starlink.

A search returns at most 10 posts, so the app runs one search for each of the last 10 days, all at the same time, and collects up to 100 posts. A run takes about a minute and a half and costs about 80 cents, most of it for X Search, which is billed at $5 per 1,000 posts it fetches, on top of token costs. To search fewer days, change DAYS at the top of src/sentiment.ts.

The shared code is in src/sentiment.ts. analyze() runs these steps and reports each one through callbacks, so the web app and the terminal app show the same progress in their own way:

1. findPosts() streams a grok-4.7 request with xSearch() limited to one day, asking for that day's most popular posts as JSON. The server_tool_call event reports the search as it runs, and the citation event collects the posts Grok cites, which is how the app drops any post that X Search didn't return. analyze() runs it for each of the last 10 days at once and removes duplicates. The client is created with retryBeforeOutput, so a search that fails before Grok starts answering is sent again, and a day whose search still fails is skipped and its error reported.
2. filterPosts() keeps the posts that show how people feel about the topic, using a low reasoning effort.
3. scoreSentiment() scores each post it kept, then the topic overall, and reports Grok's reasoning as it streams in. It uses a low effort too: at the default, Grok reasons through every post before it answers, which takes a few minutes. The schema asks for scores from -1 to 1, but the range isn't enforced, so the scores are clamped.

src/server.ts runs analyze() for the topic from the page and sends each step to it as a server-sent event. It passes an AbortSignal to every request, so closing the page or clicking Stop cancels whatever is running. public/index.html is plain HTML and JavaScript that shows those events as a dashboard, and works out the sentiment for each day from the scores of that day's posts. src/index.ts prints the same steps in the terminal.