🧠 Context-RAG

Context-RAG is not another RAG wrapper.
It's a document-understanding engine designed for real-world knowledge systems.

⚠️ Status: Beta — Actively used in production (medical RAG & enterprise docs), API stable, breaking changes documented.

⭐ If this project helps you build better RAG systems, consider giving it a star! It helps others discover this project.

⚡ 60-Second Quick Start

import { createContextRAG } from '@msbayindir/context-rag';
import { PrismaClient } from '@prisma/client';
import * as fs from 'fs';

const rag = createContextRAG({
  prisma: new PrismaClient(),
  geminiApiKey: process.env.GEMINI_API_KEY!,
});

// Ingest a PDF
await rag.ingest({
  file: fs.readFileSync('./document.pdf'),
  filename: 'document.pdf',
});

// Search with context
const results = await rag.search({
  query: 'What are the key findings?',
  mode: 'hybrid',
  useReranking: true,
});

console.log(results[0].chunk.displayContent);

📌 Requires: PostgreSQL + pgvector extension

🎯 Who is this for?

You should use Context-RAG if you...
🏗️ Are building production RAG systems (not prototypes)
📄 Process large PDFs — medical, legal, enterprise docs
🐘 Want PostgreSQL-only stack (no Pinecone, no Weaviate, no vendor lock-in)
😤 Are frustrated with context loss in traditional chunking
🔬 Need experiment tracking to compare different extraction strategies

✨ Key Features

Feature	Description
🚀 Gemini Files API	Upload PDF once, use cached URI for entire pipeline (90%+ bandwidth savings)
🧠 Contextual Retrieval	Anthropic-style context generation for each chunk (improves recall by ~49%)
🎯 Reranking	Gemini or Cohere-powered relevance reranking (reduces retrieval failure by ~67%)
🔍 Discovery Agent	AI automatically analyzes documents and suggests optimal chunking strategies
📄 Multimodal Processing	Uses Gemini Vision API to understand tables, charts, and layouts
🧪 Experiment System	A/B test different models on same document for comparison
🔎 Hybrid Search	Semantic (vector) + Keyword (full-text) search combination
🐘 PostgreSQL Native	No external vector DB needed, uses pgvector
⚡ Batch Processing	Concurrent processing with automatic retry
🛡️ Enterprise Error Handling	Correlation IDs, graceful degradation, structured logging
🔌 Dependency Injection	SOLID-compliant architecture with interface-based DI (v2.0-beta)

🏗️ Architecture

flowchart TB
    subgraph Input
        PDF[📄 PDF Document]
    end

    subgraph Discovery["🔍 Discovery Phase"]
        DA[Discovery Agent]
        DA --> |Analyzes| PS[Prompt Strategy]
        DA --> |Suggests| CS[Chunk Types]
    end

    subgraph Ingestion["📥 Ingestion Pipeline"]
        GF[Gemini Files API]
        BP[Batch Processor]
        SE[Structured Extraction]
        CR[Contextual Retrieval]
        VE[Vector Embedding]
    end

    subgraph Storage["🗄️ PostgreSQL"]
        PG[(pgvector)]
        FT[Full-Text Index]
    end

    subgraph Retrieval["🔎 Search & Retrieval"]
        HS[Hybrid Search]
        RR[Reranker]
        RR --> |Gemini or Cohere| RS[Ranked Results]
    end

    PDF --> DA
    PS --> GF
    GF --> |Cached URI| BP
    BP --> SE
    SE --> CR
    CR --> |"Adds Context"| VE
    VE --> PG
    VE --> FT
    
    Query[🔍 Query] --> HS
    HS --> PG
    HS --> FT
    PG & FT --> RR
    RS --> Response[📋 Contextual Results]

🤔 Why Contextual Retrieval?

Problem: Traditional RAG systems lose context when chunking documents. A chunk saying "The inhibitor blocks Complex IV" is meaningless without knowing it's from the "Electron Transport Chain" section.

The Anthropic Research

Anthropic's Contextual Retrieval paper showed that adding context to each chunk dramatically improves retrieval quality:

Method	Retrieval Failure Rate	Improvement
Traditional RAG	5.7%	-
+ BM25 Hybrid	4.5%	+21%
+ Contextual Retrieval	2.9%	+49%
+ Contextual + Reranking	1.9%	+67%

How Context-RAG Implements This

// Before: Raw chunk (loses context)
{
  content: "The inhibitor blocks Complex IV",
  // Where is this from? What document? What section?
}

// After: Contextual chunk (Context-RAG)
{
  content: "The inhibitor blocks Complex IV",
  contextText: "This chunk is from 'Biochemistry 101', Chapter 5: Electron Transport Chain. It describes how cyanide inhibits cytochrome c oxidase (Complex IV), stopping ATP synthesis.",
  enrichedContent: "[CONTEXT] ... [CONTENT] The inhibitor blocks Complex IV"
}

📋 Real-World Use Cases

🏥 1. Medical Education (TUS/USMLE Prep)

Scenario: Turkish medical students preparing for TUS exam with 500+ page biochemistry PDFs.

const rag = createContextRAG({
  prisma,
  geminiApiKey: process.env.GEMINI_API_KEY!,
  ragEnhancement: {
    approach: 'anthropic_contextual',  // Enable contextual retrieval
    strategy: 'llm',
    model: 'gemini-2.5-flash',
  },
});

// Discovery: AI analyzes the PDF and suggests extraction strategy
const discovery = await rag.discover({ file: pdfBuffer });

// Ingest with approved strategy
const approved = await rag.approveStrategy(discovery.id);
await rag.ingest({
  file: pdfBuffer,
  filename: 'biochemistry.pdf',
  promptConfigId: approved.id,
});

// Students can now ask contextual questions
const results = await rag.search({
  query: 'Siyanür hangi kompleksi inhibe eder?',
  mode: 'hybrid',
  useReranking: true,
});
// Returns: "Complex IV (Cytochrome c oxidase)" with full chapter context

⚖️ 2. Legal Document Analysis

Scenario: Law firms processing contracts, regulations, and case law.

// Custom extraction for legal documents
await rag.ingest({
  file: contractPdf,
  filename: 'service-agreement.pdf',
  customPrompt: `
    Extract the following from this legal document:
    - CLAUSE: Individual contract clauses with section numbers
    - DEFINITION: Defined terms and their meanings  
    - OBLIGATION: Parties' obligations and deadlines
    - LIABILITY: Liability limitations and indemnifications
  `,
});

// Search with type filtering
const liabilityClauses = await rag.search({
  query: 'limitation of liability for indirect damages',
  filters: { chunkTypes: ['LIABILITY', 'CLAUSE'] },
  useReranking: true,
});

🏢 3. Enterprise Knowledge Base

Scenario: Company onboarding with internal policies, procedures, and technical docs.

// Process multiple document types
for (const doc of ['hr-policy.pdf', 'security-guidelines.pdf', 'api-docs.pdf']) {
  const discovery = await rag.discover({ file: docs[doc] });
  const approved = await rag.approveStrategy(discovery.id);
  await rag.ingest({
    file: docs[doc],
    filename: doc,
    promptConfigId: approved.id,
    experimentId: 'knowledge-base-v1',  // Group related documents
  });
}

// Employees search across all documents
const results = await rag.search({
  query: 'What is the vacation policy for remote employees?',
  mode: 'hybrid',
});

🛡️ Enterprise Error Handling

Context-RAG implements production-grade error handling with full traceability:

Correlation IDs

Every operation is tracked with a unique correlation ID for debugging:

import { generateCorrelationId, setCorrelationId } from '@msbayindir/context-rag';

// Set correlation ID for request tracing
const correlationId = generateCorrelationId();  // crag_1737470109_abc123
setCorrelationId(correlationId);

// All logs and errors now include this ID
// [2026-01-21T18:00:00.000Z] [INFO] Starting ingestion {"correlationId":"crag_1737470109_abc123"}

Custom Error Classes

import { 
  IngestionError, 
  RerankingError, 
  ConfigurationError,
  RateLimitError 
} from '@msbayindir/context-rag';

try {
  await rag.ingest({ file: pdfBuffer, filename: 'doc.pdf' });
} catch (error) {
  if (error instanceof RateLimitError) {
    console.log(`Rate limited. Retry after ${error.retryAfterMs}ms`);
    console.log(`Correlation ID: ${error.correlationId}`);
  } else if (error instanceof IngestionError) {
    console.log(`Ingestion failed at batch ${error.batchIndex}`);
    console.log(`Retryable: ${error.retryable}`);
  }
}

Health Check

const health = await rag.healthCheck();
// {
//   status: 'healthy',
//   database: true,
//   pgvector: true,
//   reranking: { enabled: true, provider: 'gemini', configured: true }
// }

📦 Installation

npm install @msbayindir/context-rag
# or
pnpm add @msbayindir/context-rag
# or
yarn add @msbayindir/context-rag

🖥️ CLI Commands

# Initialize Context-RAG in your project (adds Prisma models to your schema)
npx @msbayindir/context-rag init

# Force overwrite existing models
npx @msbayindir/context-rag init --force

# Check setup status (Prisma models, pgvector, env variables)
npx @msbayindir/context-rag status

# Check for embedding model mismatches
npx @msbayindir/context-rag check-embeddings

# Re-index documents (useful after changing embedding models)
npx @msbayindir/context-rag reindex --concurrency 5

🛠️ Prerequisites

1. PostgreSQL with pgvector Extension

# Ubuntu/Debian
sudo apt install postgresql-15-pgvector

# macOS (Homebrew)
brew install pgvector

# Docker
docker run -e POSTGRES_PASSWORD=password -p 5432:5432 pgvector/pgvector:pg15

Then enable the extension:

CREATE EXTENSION IF NOT EXISTS vector;

2. Prisma Schema Setup

Context-RAG requires 4 models in your Prisma schema. Use our CLI to auto-add them:

# Automatically adds Context-RAG models to your schema
npx @msbayindir/context-rag init

# Then run migrations
npx prisma migrate dev --name add-context-rag

📋 Manual Setup (click to expand)

// Required: pgvector extension
generator client {
  provider        = "prisma-client-js"
  previewFeatures = ["postgresqlExtensions"]
}

datasource db {
  provider   = "postgresql"
  url        = env("DATABASE_URL")
  extensions = [vector]
}

// 4 models: ContextRagPromptConfig, ContextRagChunk, ContextRagDocument, ContextRagBatch
// Full schema: https://github.com/msbayindir/ContextRAG/blob/main/prisma/schema.prisma

3. Environment Variables

DATABASE_URL="postgresql://user:password@localhost:5432/mydb"
GEMINI_API_KEY="your-gemini-api-key"

# Optional: Cohere API key for better reranking (free tier: 10K/month)
COHERE_API_KEY="your-cohere-api-key"

🧩 Usage (Full Example)

import { createContextRAG } from '@msbayindir/context-rag';
import { PrismaClient } from '@prisma/client';

const prisma = new PrismaClient();

const rag = createContextRAG({
  prisma,
  geminiApiKey: process.env.GEMINI_API_KEY!,
  model: 'gemini-3-flash-preview',
  
  // NEW: Contextual Retrieval Enhancement
  ragEnhancement: {
    approach: 'anthropic_contextual',
    strategy: 'simple', // 'none' | 'simple' | 'llm'
  },
});

// 🔍 Discover optimal strategy
const strategy = await rag.discover({ file: './document.pdf' });
console.log(`Detected: ${strategy.documentType}`);

// ✅ Approve and create config
await rag.approveStrategy(strategy.id);

// 📥 Ingest document
const result = await rag.ingest({
  file: './document.pdf',
  onProgress: (status) => console.log(`Batch ${status.current}/${status.total}`),
});

// 🔎 Search
const results = await rag.search({
  query: 'What are the key findings?',
  mode: 'hybrid',
  limit: 10,
});

results.forEach((r) => {
  console.log(`[${r.score.toFixed(2)}] ${r.chunk.displayContent.slice(0, 100)}...`);
});

🧠 Contextual Retrieval

Context-RAG implements Anthropic's Contextual Retrieval approach using Gemini Files API.

The Problem

A chunk like "Value: 50 mg/dL" alone has no context. Searching for "Cyanide test" won't find it.

The Solution

Each chunk gets contextual information prepended:

"This chunk is from the Biochemistry Test Results table, showing 
the Cyanide test value for patient Ahmet Yılmaz. Value: 50 mg/dL"

Configuration

const rag = createContextRAG({
  // ...
  ragEnhancement: {
    approach: 'anthropic_contextual',
    strategy: 'llm',           // Best quality, uses Gemini
    skipChunkTypes: ['HEADING', 'IMAGE_REF'],
    concurrencyLimit: 5,
  },
});

Strategy	Cost	Quality Improvement
`none`	$0	Baseline
`simple`	$0	+20% (template-based)
`llm`	~$0.005/chunk	+49% (Gemini-generated)

🎯 Reranking (Advanced)

Reranking improves search relevance by re-scoring candidates using AI models. Based on Anthropic's Contextual Retrieval research, it reduces retrieval failure rate by ~67%.

How It Works

Initial Retrieval: Get top N candidates (e.g., 50) via vector similarity
Reranking: AI model scores each candidate's relevance to the query
Final Selection: Return top K (e.g., 5) based on reranked scores

Configuration

const rag = createContextRAG({
  prisma,
  geminiApiKey: process.env.GEMINI_API_KEY!,
  
  // Reranking configuration
  rerankingConfig: {
    enabled: true,
    provider: 'gemini',        // 'gemini' (free) or 'cohere' (10K/month free)
    cohereApiKey: process.env.COHERE_API_KEY, // Required if provider is 'cohere'
    defaultCandidates: 50,     // Get 50 candidates from vector search
    defaultTopK: 10,           // Return top 10 after reranking
  },
});

Per-Query Reranking

const results = await rag.search({
  query: 'metabolizma ve enerji üretimi',
  limit: 5,
  useReranking: true,          // Enable reranking for this query
  rerankCandidates: 50,        // Get 50 candidates, rerank to top 5
});

// Results include reranking metadata
results.forEach(r => {
  console.log(`Score: ${r.score}`);
  console.log(`Reranked: ${r.explanation?.reranked}`);
  console.log(`Original rank: ${r.explanation?.originalRank}`);
});

Provider Comparison

Provider	Cost	Quality	Best For
Gemini	Free (uses existing quota)	Good	Cost-sensitive, general use
Cohere	Free tier: 10K/month	Excellent	Multilingual, production

🎯 Custom Prompt / Filtered Extraction (Advanced)

Extract only specific content types without going through the Discovery flow:

// Extract ONLY specific types with custom prompt
const result = await rag.ingest({
  file: './book.pdf',
  customPrompt: `
    Extract ONLY these content types:
    - TEXT: Normal paragraphs
    - QUESTION: Multiple choice questions
    - LIST: Bulleted or numbered lists
    - TABLE: Data tables

    SKIP these types:
    - HEADING, CODE, QUOTE, IMAGE_REF
  `,
  // Context enrichment only for TEXT chunks (cost optimization)
  // Configure via ragEnhancement.skipChunkTypes
});

Configuration for Selective Context Enrichment

const rag = createContextRAG({
  prisma,
  geminiApiKey: process.env.GEMINI_API_KEY!,
  
  ragEnhancement: {
    approach: 'anthropic_contextual',
    strategy: 'llm',
    // Only TEXT chunks get context enrichment
    // Other types (TABLE, LIST, QUESTION) are extracted but not enriched
    skipChunkTypes: ['HEADING', 'IMAGE_REF', 'TABLE', 'CODE', 'QUOTE', 'MIXED', 'QUESTION', 'LIST'],
  },
});

// PromptConfig is auto-created when using customPrompt
await rag.ingest({
  file: './document.pdf',
  customPrompt: 'Your custom extraction instructions...',
});

Note: When using customPrompt without promptConfigId, the system automatically creates a PromptConfig for you.

⚙️ Configuration

Context-RAG is highly configurable. Below is the complete list of all available options.

const rag = createContextRAG({
  // ============================================
  // CORE CONFIGURATION (Required)
  // ============================================

  /** Your initialized Prisma client instance */
  prisma: prismaClient,

  /** Gemini API Key (Required for generation and default embeddings) */
  geminiApiKey: process.env.GEMINI_API_KEY!,

  // ============================================
  // MODEL SELECTION
  // ============================================

  /** 
   * Main LLM model for generation, orchestration, and RAG enhancement.
   * Default: 'gemini-1.5-pro'
   */
  model: 'gemini-1.5-pro', // Options: 'gemini-1.5-flash', 'gemini-2.0-flash-exp', etc.

  /**
   * Configuration for the LLM generation (temperature, tokens, etc.)
   */
  generationConfig: {
    temperature: 0.3,        // Creativity (0.0 - 1.0). Lower is more deterministic.
    maxOutputTokens: 8192,   // Maximum length of the generated response.
  },

  // ============================================
  // EMBEDDING PROVIDER (Optional)
  // ============================================

  /**
   * Choose your embedding provider.
   * Default: Uses Gemini 'text-embedding-004'
   */
  embeddingProvider: {
    // Provider: 'gemini' | 'openai' | 'cohere'
    provider: 'openai', 

    // Model name (specific to the provider)
    model: 'text-embedding-3-small',

    // API Key (if different from geminiApiKey)
    apiKey: process.env.OPENAI_API_KEY,
  },

  // ============================================
  // SYSTEM CONFIGURATION
  // ============================================

  /**
   * Batch processing settings for ingestion.
   * Adjust these based on your API rate limits.
   */
  batchConfig: {
    pagesPerBatch: 15,       // How many pages to process in one go (Default: 15)
    maxConcurrency: 3,       // How many batches to run in parallel (Default: 3)
    maxRetries: 3,           // Retry failed batches (Default: 3)
    retryDelayMs: 1000,      // Initial delay before retry (Default: 1000ms)
    backoffMultiplier: 2,    // Exponential backoff factor (Default: 2)
  },

  /**
   * Settings for splitting text into vector chunks.
   */
  chunkConfig: {
    maxTokens: 500,          // Maximum size of a single chunk (Default: 500)
    overlapTokens: 50,       // Overlap between chunks to preserve continuity (Default: 50)
  },

  /**
   * API Rate Limiting protection.
   */
  rateLimitConfig: {
    requestsPerMinute: 60,   // Max RPM allowed (Default: 60)
    adaptive: true,          // Automatically slow down if 429 errors occur (Default: true)
  },

  /**
   * System logging configuration.
   */
  logging: {
    level: 'info',           // 'debug' | 'info' | 'warn' | 'error'
    structured: true,        // Use JSON format for logs (Best for production tools like Datadog/CloudWatch)
  },

  // ============================================
  // ADVANCED FEATURES
  // ============================================

  /**
   * Reranking improves search relevance by re-scoring results.
   */
  rerankingConfig: {
    enabled: true,           // Enable automatic reranking (Default: false)
    provider: 'cohere',      // 'gemini' or 'cohere' (Cohere is recommended for best results)
    cohereApiKey: process.env.COHERE_API_KEY, // Required if provider is 'cohere'
    defaultCandidates: 50,   // Retrieve top 50 from Vector DB...
    defaultTopK: 10,         // ...and return top 10 after reranking.
  },

  /**
   * RAG Enhancement (Contextual Retrieval).
   * Adds context to chunks before embedding them.
   */
  ragEnhancement: {
    // Approach: 'anthropic_contextual' (Recommended) or 'none'
    approach: 'anthropic_contextual',

    // Strategy: 'llm' (Best Quality) or 'simple' (Template based)
    strategy: 'llm',

    // Model to use for generating context (Optional, defaults to main model)
    // Tip: Use a cheaper model here (e.g., 'gemini-1.5-flash') to save costs.
    model: 'gemini-1.5-flash',

    // Prompt used to generate context (Optional, has good default)
    contextPrompt: 'Situate this chunk within the document...',

    // Don't waste tokens generating context for these types
    skipChunkTypes: ['HEADING', 'IMAGE_REF', 'CODE'], 
  },

  /** 
   * Enable Structured Output (JSON Schema) for reliable parsing.
   * Disable only if you are using a model that doesn't support it well.
   * Default: true 
   */
  useStructuredOutput: true,

  /**
   * Custom Chunk Type Mapping.
   * Map your custom extraction types to system types for proper handling.
   */
  chunkTypeMapping: {
    'RECIPE': 'TEXT',        // Treat 'RECIPE' as normal text
    'INGREDIENT_LIST': 'LIST', // Treat 'INGREDIENT_LIST' as a list
    'NUTRITIONAL_INFO': 'TABLE' // Treat 'NUTRITIONAL_INFO' as a table
  }
});

📚 API Reference

Discovery

const strategy = await rag.discover({
  file: pdfBuffer,
  documentTypeHint: 'Medical',
});

await rag.approveStrategy(strategy.id);

Ingestion

const result = await rag.ingest({
  file: pdfBuffer,
  filename: 'report.pdf',
  documentType: 'Medical',
  experimentId: 'exp_v1',  // For A/B testing
  skipExisting: true,
  onProgress: (status) => console.log(status),
});

Search

const results = await rag.search({
  query: 'medication interactions',
  mode: 'hybrid',
  limit: 20,
  minScore: 0.5,
  filters: {
    documentTypes: ['Medical'],
    chunkTypes: ['TABLE', 'TEXT'],
  },
  typeBoost: {
    TABLE: 1.5,
  },
});

📤 Publishing to npm

If you want to publish your own fork:

# 1. Login to npm
npm login

# 2. Build the package
pnpm build

# 3. Publish (first time)
npm publish --access public

# 4. Publish update
npm version patch  # or minor/major
npm publish

🧪 Development

# Install dependencies
pnpm install

# Build
pnpm build

# Lint
pnpm lint

# Type check
pnpm typecheck

# Run demo
pnpm demo

🤝 Contributing

Contributions are welcome! Here's how to get started:

Getting Started

Fork the repository
Clone your fork: git clone https://github.com/YOUR_USERNAME/ContextRAG.git
Install dependencies: pnpm install
Create a branch: git checkout -b feature/amazing-feature

Making Changes

Make your changes
Run linting: pnpm lint
Run build: pnpm build
Test your changes locally

Submitting a PR

Commit your changes: git commit -m 'feat: add amazing feature'
Push to your fork: git push origin feature/amazing-feature
Open a Pull Request

Commit Convention

We use Conventional Commits:

feat: New feature
fix: Bug fix
docs: Documentation only
refactor: Code change that neither fixes nor adds
test: Adding tests
chore: Build process or auxiliary tool changes

Code Style

TypeScript strict mode
ESLint + Prettier
Meaningful variable/function names
JSDoc comments for public APIs

📁 Project Structure

context-rag/
├── src/
│   ├── context-rag.ts       # Main facade class
│   ├── context-rag.factory.ts # DI Factory (v2.0-beta)
│   ├── engines/             # Discovery, Ingestion, Retrieval
│   ├── enhancements/        # RAG Enhancement handlers
│   │   └── anthropic/       # Anthropic Contextual Retrieval
│   ├── services/            # Gemini API, PDF Processor
│   ├── providers/           # Embedding providers (Gemini, OpenAI, Cohere)
│   ├── database/            # Prisma repositories
│   ├── config/              # Templates & constants
│   ├── types/               # TypeScript types & interfaces
│   ├── utils/               # Logger, Retry, RateLimiter
│   └── errors/              # Custom error classes
├── examples/                # Demo scripts
├── tests/                   # Unit & integration tests
├── prisma/                  # Reference schema
└── dist/                    # Built output

🔄 Migration Guide (v1.x → v2.0-beta)

Breaking Change: Factory Pattern

v2.0-beta introduces proper Dependency Injection. The new ContextRAG() constructor now requires dependencies.

Before (v1.x):

import { ContextRAG } from '@msbayindir/context-rag';

const rag = new ContextRAG({
  prisma,
  geminiApiKey: 'your-key',
});

After (v2.0-beta):

import { createContextRAG } from '@msbayindir/context-rag';

const rag = createContextRAG({
  prisma,
  geminiApiKey: 'your-key',
});

Custom Engine Injection (Advanced)

v2.0-beta allows injecting custom engines for advanced use cases:

import { ContextRAG, IngestionEngine } from '@msbayindir/context-rag';

// Create custom engine
class MyIngestionEngine extends IngestionEngine {
  async ingest(options) {
    console.log('Custom logic!');
    return super.ingest(options);
  }
}

// Inject via constructor
const rag = new ContextRAG(config, {
  ingestionEngine: myCustomEngine,
  retrievalEngine,
  discoveryEngine,
  repos: { promptConfig, document, chunk },
});

📄 License

🙏 Acknowledgments

Anthropic for the Contextual Retrieval research
Google for Gemini API and Files API
pgvector for PostgreSQL vector support

Name		Name	Last commit message	Last commit date
Latest commit History 63 Commits
.github/workflows		.github/workflows
docs		docs
examples		examples
prisma		prisma
scripts		scripts
src		src
tests		tests
.env.example		.env.example
.gitignore		.gitignore
.npmignore		.npmignore
.prettierrc		.prettierrc
LICENSE		LICENSE
README.md		README.md
debug-schema.ts		debug-schema.ts
eslint.config.js		eslint.config.js
package.json		package.json
pnpm-lock.yaml		pnpm-lock.yaml
tsconfig.json		tsconfig.json
tsup.config.ts		tsup.config.ts
vitest.config.ts		vitest.config.ts

Folders and files

Latest commit

History

Repository files navigation

🧠 Context-RAG

⚡ 60-Second Quick Start

🎯 Who is this for?

✨ Key Features

🏗️ Architecture

🤔 Why Contextual Retrieval?

The Anthropic Research

How Context-RAG Implements This

📋 Real-World Use Cases

🏥 1. Medical Education (TUS/USMLE Prep)

⚖️ 2. Legal Document Analysis

🏢 3. Enterprise Knowledge Base

🛡️ Enterprise Error Handling

Correlation IDs

Custom Error Classes

Health Check

📦 Installation

🖥️ CLI Commands

🛠️ Prerequisites

1. PostgreSQL with pgvector Extension

2. Prisma Schema Setup

3. Environment Variables

🧩 Usage (Full Example)

🧠 Contextual Retrieval

The Problem

The Solution

Configuration

How It Works

Configuration

Per-Query Reranking

Provider Comparison

Configuration for Selective Context Enrichment

⚙️ Configuration

📚 API Reference

Discovery

Ingestion

Search

📤 Publishing to npm

🧪 Development

🤝 Contributing

Getting Started

Making Changes

Submitting a PR

Commit Convention

Code Style

📁 Project Structure

🔄 Migration Guide (v1.x → v2.0-beta)

Breaking Change: Factory Pattern

Custom Engine Injection (Advanced)

📄 License

🙏 Acknowledgments

About

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases 7

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages