Reduce your AI coding assistant token usage by 5x–10x. A local, privacy-first tool that tracks file changes, logs prompts, and generates compact context summaries so your AI always has exactly what it needs — nothing more.

Author: Muhammad Faizan GitHub: github.com/faizzyhon Instagram: instagram.com/faizzyhon

Every time you start a new AI session, you re-paste hundreds or thousands of tokens of context. Token Buster solves this:

• Auto-watches your project directory for file changes after each AI edit
• Logs every prompt/response with before-and-after diffs and token counts
• Generates a compact context log — a single copy-pasteable block that gives the AI 100% context in 10–20% of the tokens
• Dashboard to track usage, costs, and savings across all your projects

token_buster/
├── backend/
│   └── app.py           Flask REST API (12 endpoints)
├── core/
│   ├── tokenizer.py     tiktoken / HuggingFace / fallback
│   ├── watcher.py       Watchdog-based file monitor
│   └── summarizer.py    Context log generator
├── db/
│   └── database.py      SQLite (WAL mode, 4 tables)
└── frontend/
    └── src/
        ├── App.js
        ├── api.js
        └── components/
            ├── Sidebar.js
            ├── Dashboard.js
            ├── ProjectView.js
            ├── TokenCounter.js
            ├── LogEditor.js
            └── ContextLogModal.js

bash start.sh

start.bat

# 1. Install Python deps
pip install -r requirements.txt

# 2. Build the React frontend
cd token_buster/frontend
npm install
npm run build
cd ../..

# 3. Launch
python main.py --port 5000

Open: http://localhost:5000

• Global token usage + savings across all projects
• Per-project stats: input/output tokens, prompts, file changes
• One-click start/stop file watching

• Real-time token efficiency bar
• Prompt log timeline with event types (manual, ai-edit, refactor, debug)
• Auto-tracked file changes with token counts

• Auto-generated copy-paste block with all recent changes
• Shows token count before vs. after compression
• Add custom notes to inject into context

• Count tokens for any text with cost estimates
• Side-by-side diff: compare before/after an AI edit
• Supports: Claude 3.5 Sonnet, Claude 3 Opus, Haiku, GPT-4o, etc.

• Edit prompt summaries and response summaries
• Update "after content" to capture what changed
• Full CRUD on all prompt records

Before Token Buster:
  You paste: 3,200 tokens of file contents + history

After Token Buster:
  Context log: 320 tokens — same information, 10x less

Monthly savings (100 sessions/day):
  ~28,800,000 tokens → ~$86 saved on Claude 3.5 Sonnet

MIT © 2025 Muhammad Faizan — github.com/faizzyhon