Turn Cosmos DB into
your RAG engine
Serverless RAG indexing on Cosmos DB. Scales linearly. Goes global.
Drop the limitations of Azure AI Search. RAG DB turns Cosmos DB into a full-featured vector database with 30+ file type ingestion, four search modes, and the global scale Cosmos was built for.
Any File (30+ types)
RAG.DB Engine
Key Vault
Managed Identity
App Insights
Container Apps
Cosmos DB
The Problem
Cosmos DB is a great vector database.
But who builds the ingestion pipeline?
You chose Cosmos DB for its global scale and vector search. But getting your data in — that's the hard part.
0%
Unstructured data
PDFs, DOCX, audio, video, images, emails — you need high-fidelity Markdown extraction that preserves tables, layouts, and structure from the originals.
0
Built-in sync
Files get added, edited, and deleted in Blob Storage. Without an event-driven pipeline, your index drifts out of sync silently.
0
Search modes needed
Vector, hybrid, full-text, and semantic search. You need all four for production RAG — plus query rewriting for better recall.
N
Chunking strategies
Naive splitting destroys context. Great extraction is wasted without intelligent chunking that respects headings, tables, and section boundaries.
0+
Container limit
Cosmos DB caps at 500 containers per account. At scale, you need automatic provisioning across multiple accounts.
24/7
Monitoring & recovery
Stuck jobs, failed conversions, stale indexes. Without self-healing, your pipeline needs constant babysitting.
How It Works
From upload to search
in minutes
Any File
30+ types
Event Grid
Auto-detect
RAG.DB Engine
Chunk + Embed
Cosmos DB
Vector + Full-Text
Query API
4 search modes
Any File
30+ types
Event Grid
Auto-detect
RAG.DB Engine
Chunk + Embed
Cosmos DB
Vector + Full-Text
Query API
4 search modes
Powered by Azure
Key VaultManaged IdentityApp InsightsContainer Apps
Service Bus
Container RegistryKey Capabilities
The full RAG pipeline, out of the box
4 Search Modes
Vector, hybrid, full-text, and semantic — all included in every index. Plus AI-powered query rewriting that expands queries for better recall.
Per-Index Isolation
Every index gets its own Container App, Service Bus topic, and Cosmos DB container. Fully isolated — zero noisy neighbors, independent scaling.
Event-Driven Ingestion
Fully serverless. Drop files into Blob Storage, Event Grid triggers processing, Container Apps scale to zero when idle. Pay only for what you use.
30+ File Types
PDF, DOCX, audio, images, email, archives — converted via the open-source markitdown-pro library with OCR and transcription.
Zero-Secret Security
Managed identity, Key Vault, dual API keys with rotation, user delegation SAS tokens. No credentials in code.
Multi-Cosmos Scale-Out
Cosmos DB gives you global distribution out of the box. RAG DB scales linearly across accounts — no search unit limits, no partition caps.
Self-Healing Monitoring
CronJob monitors every 5 minutes. Detects stale indexes, recovers stuck runs, and retries failed files automatically.
Observability Built-In
Application Insights telemetry, custom histograms for indexing and query performance, Cosmos DB RU tracking per query.
Optimistic Concurrency
ETag-based concurrency control throughout the pipeline. No race conditions, no data corruption, even at scale.
Powered by markitdown-pro
Our open-source conversion library turns 30+ file types into clean Markdown. It cascades through multiple extraction methods until it gets the best possible content.
Cascading Conversion Pipeline
Each file goes through up to 4 extraction methods. If one method can't extract enough content, the next one takes over automatically.
MarkItDown
Microsoft's base converter handles common formats natively
Unstructured.io
Advanced document parsing for complex layouts and tables
Azure Document Intelligence
OCR and structured extraction for scanned documents
GPT-4o-mini Vision
AI-powered OCR as a last resort for difficult content
30+ Supported File Types
From PDFs and Office docs to audio transcription, email extraction, and OCR on scanned images — one library handles it all.
PDF, DOCX, PPTX, XLSX, ODT, EPUB, RTF
PNG, JPG, WEBP, HEIC, TIFF, SVG, GIF
WAV, OGG, FLAC, M4A, AAC, WMA, OPUS
EML, PST (with attachment extraction)
CSV, JSON, XML, TSV
Python, JS, TS, Java, C#, Go, Rust, ...
HTML, Markdown, reStructuredText
ZIP (recursive processing)
MIT License · Python ≥ 3.12 · Published by Kinetic Solutions Group
See It In Action
One API to index and search
30+
File types
4
Search modes
∞
Global scale
Ready to turn Cosmos DB into your RAG engine?
Drop your files, RAG DB handles the rest — extraction, chunking, embeddings, sync, and search. Serverless, globally scalable, built on Cosmos DB.