TokenSense 🔍

TokenSense is a powerful cost-optimization platform that analyzes LLM API usage and Cloud Resource logs to detect waste patterns and generate actionable, high-saving recommendations.

📋 Progress Status

• [x] Log Aggregation System: Grouping disparate LLM and Cloud logs into actionable metrics.
• [x] Deterministic Rules Engine: High-fidelity detection of model misuse, long prompts, and over-provisioned resources.
• [x] Recommendation Engine: Mapping insights to dollar-saving actions with effort and confidence ratings.
• [x] CLI Reporting Tool: Instant terminal-based reporting for quick audits.
• [x] LLM Enrichment: Adding intelligent summaries and context to cost insights.
• [x] React Web Dashboard: High-fidelity visual analytics for cost and waste trends.
• [ ] Real-time API Integration: (Next Step) Moving from mock data to live production log ingestion.
• [ ] Automated Remediation: (Future) One-click fixes for identified cloud over-provisioning.

🚀 Getting Started

Prerequisites

• Node.js >= 18
• npm

Installation

git clone https://github.com/Siddhant2713/TokenSense.git
cd TokenSense
npm install

Running the Dashboard

npm run dev

Running the CLI

npm run cli

🛠️ Tech Stack

• Frontend: React 19, Vite, Recharts (for visual analytics)
• Backend/Logic: TypeScript
• CLI: Standard console output + tsx runtime
• Design: Premium CSS with a focus on dark-mode aesthetics

🔍 Detection Rules

| Rule | Description | Saving Potential | | :--- | :--- | :--- | | Model Misuse | Detects expensive models used for low-complexity tasks. | High | | Long Prompts | Identifies excessive prompt context that could be optimized via RAG. | Medium | | RAM Over-Provision | Flags cloud resources with significantly higher RAM than required. | High | | Usage Spikes | Anomalous burst detection to catch looping scripts or bugs. | High |

Built for the Mela VC Cost Optimization Hackathon. TokenSense helps you stop leaving money on the table.