ChatLens

ChatLens analyses a chat export from WhatsApp, Instagram or Messenger and shows who talks more, when you talk, how fast you reply, how long your streaks run and a few playful scores. All of the work happens in your browser. The site is static, has no backend, and your chat file is never uploaded anywhere.

Live demo: https://chatlens.somansh.in Sample analysis without uploading anything: https://chatlens.somansh.in/?demo

What it analyses

• Message, word and emoji counts per person, words per message, and who starts conversations (a message after a gap of more than 4 hours counts as a new conversation).
• Activity by hour of day, busiest day, and the longest run of consecutive days with messages.
• Average response time.
• Sentiment per person (positive, neutral, negative share of messages).
• Signature phrases (words and two-word pairs each person uses most) and top emojis.
• Call statistics, when the export contains calls (Instagram and Messenger).
• Light-hearted scores: a vibe check, a communication style label, an energy match based on peak hours, and a consistency score with a day-by-day heatmap.

Privacy

• The app is a static Next.js site. There is no server code and no database.
• There are no analytics, trackers or third-party scripts.
• Nothing is stored. The chat is read into memory, analysed, and gone when you close or reload the tab. There is no localStorage, cookie or upload.
• The site sends a Content-Security-Policy with connect-src 'none', so the browser refuses the page's own network calls (fetch, XHR, WebSocket and beacon). It also sets form-action 'none' and default-src 'self'. The policy is in next.config.mjs. It is applied to production builds only, because the dev server needs a WebSocket for hot reload.

You do not have to take this on trust:

1. Offline test: load the page, disconnect from the network, then analyse a chat. It works the same.
2. Network tab: open your browser's developer tools, go to Network, and analyse a chat. No request is made while or after you pick a file.
3. CSP header: in the Network tab, click the document request and read the Content-Security-Policy response header, or run curl -I https://chatlens.somansh.in.

Supported formats

Platform	File	Notes
WhatsApp	.txt export	Android style (15/10/25, 14:17 - Name: text) and iPhone style ([10/15/25, 2:17:39 PM] Name: text). Multi-line messages are joined; system notices and "media omitted" lines are skipped.
WhatsApp	.json	Accepted by the parser.
Instagram	message_*.json	Upload every message_*.json file from the chat folder; they are merged.
Messenger	message_*.json	Same as Instagram.
Discord		Listed in the UI as coming soon. Not usable yet.

At least 10 messages are needed to produce an analysis.

How to export a chat

• WhatsApp: open the chat, tap the menu (More), choose Export chat, choose Without media, and save the .txt file.
• Instagram: Settings, then Accounts Center, then Your information and permissions, then Download your information. Choose the Messages data, select JSON as the format, and download. Unzip it and find the chat's folder under messages/inbox/.
• Messenger: Settings, then Accounts Center, then Your information and permissions, then Download your information. Choose Messages, select JSON, download, unzip, and find the chat's folder under messages/inbox/.

Menu names change from time to time; the in-app hint for each platform is the shortest version. For Instagram and Messenger, upload all message_*.json files in the chat folder, since long chats are split across several files.

Run it

Requires Node.js 20 or newer.

npm install
npm run dev          # development server on http://localhost:3000
npm run build        # production build
npm run start        # serve the production build
npm run typecheck    # tsc --noEmit
npm run check        # parser self-check

npm run check runs lib/process-chat.check.ts, which feeds sample Android and iPhone WhatsApp exports through the parser and asserts on the result. The production build also fails on type errors.

How it works

• Multi-file exports (Instagram, Messenger) are de-duplicated across files only; a message repeated within one file is kept.
• Parsing (lib/process-chat.ts): each platform has a parser that turns the file into a flat list of { sender, content, timestamp }. WhatsApp text is matched line by line with a regular expression that accepts both export styles and several date separators. Instagram and Messenger JSON is read from the messages array; call records are pulled out separately.
• Sentiment: a word list. Each message is lowercased and its words are checked against a positive list and a negative list (plus a few emoticons). More positive hits makes the message positive, more negative hits makes it negative, otherwise neutral.
• Streaks: messages are grouped by calendar day in your browser's time zone, and the longest run of consecutive days with at least one message is reported.
• Response time: for consecutive messages from different people (all senders, calls included), the gap counts as a reply if it is longer than 0 and under 60 minutes, and the average is taken. Longer gaps are treated as a new conversation, not a slow reply. Only the first 2,000 messages are used, and if no gap qualifies it shows 5 minutes. When a chat has calls, response time is still shown; call time is in the call panel.
• Signature phrases: words longer than two letters that are not stop words, plus adjacent word pairs, counted per person. Large chats are sampled (every 3rd, 5th or 10th message depending on size), and a phrase needs at least 3 occurrences.
• Scores: the vibe check combines message balance (30%), positive sentiment share (50%) and emoji use (up to 20%), averaged over the whole chat. The other labels come from simple thresholds on the same numbers.

Limitations

• Sentiment is a word list. It does not understand context, sarcasm, slang it has not been given, or any language other than English.
• The scores and labels are for fun. They are not a measure of a relationship.
• Only the first two participants (the two with the most messages) are compared. In a group chat, everyone else is left out of the head-to-head views.
• Signature phrases are mostly single words, and sampling means counts are approximate on large chats.
• Response time ignores gaps of an hour or more and looks only at the start of the chat.
• Times are shown in your browser's time zone.
• Emoji counting uses fixed Unicode ranges and misses some newer emoji.

Tech

Next.js 16, React 19, Tailwind CSS v4, lucide-react icons. Charts are drawn by hand in SVG and CSS, so there is no charting library.

Licence

No licence has been chosen yet, so by default all rights are reserved.