About

A tool that turns chat text into short AI comments. Use the results as a starting point for reflection.

1. What Is Chat Analyzer? (Purpose & Vision)

Chat Analyzer (talgok.com) is a web utility that takes exported messenger chat logs (.txt) from apps like KakaoTalk, LINE, or WhatsApp and summarizes each participant’s speech habits and interaction patterns into readable nicknames and concise comments. In text-only conversations without vocal tone or facial expressions, it is designed to help you step back and review habits you might not notice—such as rapid-fire questioning, repeated one-word replies, or deferring decisions—as well as each person’s actual contributions from an objective third-party perspective.

Mild Mode and Spicy Mode act as two interpretive lenses for the exact same conversation: one highlights cooperation, consideration, and positive effort, while the other playfully critiques blunt or repetitive communication habits. This tool does not provide medical or psychological diagnoses, nor does it calculate hidden feelings or attraction scores; it serves purely as a starting point for calmly reflecting on the original chat.

2. Text Preprocessing Engine (prepareChat) & Regex Parsing

Raw chat export files contain metadata that clutters analysis, including date dividers, export timestamps, group-chat join/leave notices, and per-message timestamps. Chat Analyzer’s in-browser preprocessing engine (prepareChat) uses regular expressions to automatically recognize KakaoTalk desktop bracket format ([Name] [2:30 PM] Message), KakaoTalk mobile date-time format (2026. 10. 5. 2:30 PM, Name : Message), tab-delimited exports, and standard colon-delimited text.

During preprocessing, only system notices and leading message timestamps are stripped—actual times or dates mentioned inside the conversation itself (e.g., “Let’s meet at 14:30 tomorrow”) are preserved intact. To prevent distortion when someone splits one thought into multiple rapid bubbles, consecutive messages from the same speaker are merged into a single Turn, and up to the latest 3,000 turns (within 20,000 characters) are selected for analysis.

3. Dual Prompt Architecture (Mild / Spicy) & Guardrails

The cleaned conversation text is sent via Cloudflare Pages Functions to the Google Gemini model. System instructions strictly isolate the uploaded chat log as Untrusted Data rather than executable commands to defend against prompt injection.

The Mild Mode prompt directs the model to identify observable cooperation, care, concrete proposals, and responsiveness in a warm tone. The Spicy Mode prompt highlights noticeable habits like one-sided questioning or terse replies with wit and candor. Both modes enforce strict output guardrails prohibiting personal insults or fabricated metrics, percentages, attachment styles, or MBTI diagnoses.

4. Zero-Retention Policy & Privacy Guidelines

Chat Analyzer does not operate any database to store or collect uploaded chat logs or generated AI results. Files passed through the mobile Web Share Target API are also deleted from the browser’s temporary cache immediately after being read.

However, clicking the analyze button transmits the preprocessed text through Cloudflare servers to an external AI provider (Google Gemini API). Before uploading, you must manually remove or replace any personally identifiable information—such as real names, phone numbers, home or work addresses, bank account numbers, and passwords—with pseudonyms like “A” and “B”.

5. Editorial Standards for Magazine & Case Studies

All guides across our Magazine (20 articles) and Case Studies (8 scenarios) are authored and edited by the Chat Analyzer editorial team (The HIST) to help readers interpret digital communication more clearly. Any sample dialogues shown in our guides are fictional reconstructions created for illustration—we never quote private user chats or present unverified statistics as fact.

Every guide follows a four-step framework: (1) observable text actions in the dialogue, (2) plausible interpretations depending on context, (3) what text alone cannot prove, and (4) concrete request phrasing to align expectations.

6. Operator Information & Technical Support

Chat Analyzer is operated and maintained by The HIST. For file-parsing issues, browser compatibility inquiries, privacy questions, or factual correction requests regarding our guides, please contact our official support email at [email protected].

Reported technical issues and editorial feedback are reviewed and reproduced by our team before updates are applied, with each document’s last-updated date transparently maintained.

Related Pages

How to Use · FAQ · Contact