🛡️ methara

The high-performance, local-first utility belt for AI Agents.

methara is a lightweight, zero-dependency (at runtime) TypeScript library designed to provide AI Agents with the essential tools they need to process human language locally, securely, and with extreme speed.

📖 Documentation

For full API reference, advanced guides, and integration examples, visit our official documentation:

👉 methara.gitbook.io/methara-docs

🚀 Quick Start

Installation

npm install methara

Basic Usage

import { Intent, Entities, Guard } from 'methara';

const input = "Hey @alice, remind me to check the ORD-12345 status tomorrow at 2pm";

// 1. Sanitize input (Mask PII)
const cleanInput = Guard.maskPII(input);

// 2. Detect Intent
const intent = Intent.analyze(cleanInput); 
// { type: 'task', action: 'reminder', confidence: 0.85 }

// 3. Extract Entities
const data = Entities.extract(input);
// { date: 2026-05-09..., time: '14:00', mentions: ['@alice'], trackingId: 'ORD-12345' }

🧩 Modules

🧠 Intent Detection

Analyze user messages to determine what they want to do without calling an LLM.

• Actions: create, delete, view, update, complete, reminder.
• Scope: single, all, team.

🔍 Entity Extraction

Extract structured data from natural language.

• Dates & Times: Relative (tomorrow, next Friday) and absolute formats.
• Mentions: Extracts @usernames.
• Tracking IDs: Detects common formats like ORD-123, TKT-456.
• Quantities: Converts word-numbers ("three") to integers (3).

🛡️ Guard (Security & PII)

Protect your agent from prompt injection and sensitive data leaks.

• PII Masking: Emails, Phone numbers, Credit Cards, SSNs, IP Addresses.
• Injection Protection: Detects common "jailbreak" and "ignore previous instructions" phrases.
• Profanity Filter: Localized check for toxic language.

📊 User Patterns

Understand how your users communicate over time.

• Urgency Level: Detects high, medium, or low based on tone and keywords.
• Style Analysis: Identifies formal vs casual and detailed vs concise communication.
• Preferences: Tracks time-of-day and task-style preferences.

⚙️ Text Processor

Optimize your context window and token usage.

• Shrink: Removes stopwords and normalizes whitespace (reduces tokens by 20-40%).
• Token Counter: Accurate character-based token approximation.
• Signature Removal: Strips email signatures and boilerplate.

🛠️ Advanced Configuration

methara is built with ESM and TypeScript from the ground up, ensuring full compatibility with modern Node.js environments (v18+).

import { Processor } from 'methara';

const longText = "Very long conversation history...";
const optimized = Processor.shrink(longText, {
  removeStopwords: true,
  removeSignature: true
});

📄 License

MIT © 2026