π¦ chatpack
Feed your chat history to LLMs. Compress exports 13x with CSV format.
Platforms: Windows β’ macOS β’ Linux
The Problem
You want to ask Claude/ChatGPT about your conversations, but:
- Raw exports are 80% metadata noise
- JSON structure wastes tokens on brackets and keys
- Context windows are expensive
The Solution
βββββββββββββββββββ ββββββββββββ βββββββββββββββββββ
β Telegram JSON β β β β Clean CSV β
β WhatsApp TXT β βββΆ β chatpack β βββΆβ Ready for LLM β
β Instagram JSON β β β β 13x less tokens β
βββββββββββββββββββ ββββββββββββ βββββββββββββββββββ
Real Numbers
| Format | Input (Telegram JSON) | Output | Savings |
|---|---|---|---|
| CSV | 11.2M tokens | 850K tokens | 92% (13x) π₯ |
| JSONL | 11.2M tokens | 1.0M tokens | 91% (11x) |
| JSON | 11.2M tokens | 1.3M tokens | 88% (8x) |
π‘ Use CSV for maximum token savings. JSONL is good for RAG pipelines. JSON keeps full structure but wastes tokens.
Use Cases
π¬ Chat with your chat history
# Paste into ChatGPT: "Based on this conversation, what did we decide about...?"
π Build RAG pipeline
# Each line = one document with timestamp for vector DB
π Analyze conversations
# Filter and export specific messages
Features
- π Fast β 20K+ messages/sec
- π± Multi-platform β Telegram, WhatsApp, Instagram
- π Smart merge β Consecutive messages from same sender β one entry
- π― Filters β By date, by sender
- π Formats β CSV (13x compression), JSON, JSONL (for RAG)
Installation
Pre-built binaries (recommended)
| Platform | Download |
|---|---|
| Windows | chatpack-windows-x64.exe |
| macOS (Intel) | chatpack-macos-x64 |
| macOS (Apple Silicon) | chatpack-macos-arm64 |
| Linux | chatpack-linux-x64 |
Via Cargo
Quick Start
# Telegram
# WhatsApp
# Instagram
Output: optimized_chat.csv β ready to paste into ChatGPT/Claude.
Documentation
| Guide | Description |
|---|---|
| π€ Export Guide | How to export from Telegram, WhatsApp, Instagram |
| π Usage Guide | All commands, flags, filters, formats |
| π Benchmarks | Performance stats and compression metrics |
| π§ͺ Stress Testing | Generate toxic data and run stress tests |
Quick Reference
# Output formats
# Filters
# Metadata
Technical Details
| Source | Format | Features |
|---|---|---|
| Telegram | JSON | IDs, timestamps, replies, edits |
| TXT | Auto-detect locale (US/EU/RU), multiline | |
| JSON | Mojibake fix, empty message filter |
Performance
| Metric | Value |
|---|---|
| Speed | 20-50K messages/sec |
| CSV compression | 13x (92% token reduction) |
| Tested file size | 500MB+ |