Multimodal Embedding Preview (Alpha)
Try more advanced controls ->
Trusted by builders at
Large token limit
Great for long context retrievals with support for 8192 input tokens
Smaller vector size
Highly optimized vector size of 768 dimensions for efficient storage and retrieval
Native language support
Supports over 100+ languages natively with MTEB score of 82.11
Native media support
Custom encoding layers for image, pdf, CSVs, code & audio
Optimized
Optimized for retrieval tasks and similarity search
Upgradable
Encoding layers can be upgraded without breaking existing generated vectors
Integrate Multimodal Embedding on any platform
JavaScript
Python
PHP
Ruby
Go
Java
Swift
Dart
Kotlin
C#
cURL
npm i jigsawstack
Multimodal Embedding use cases
4 ways our customers use JigsawStack's Multimodal Embedding to build applications
Enterprise RAGs
Build RAGs for your enterprise with support for multiple media types and languages
Recommendation engines
Build recommendation engines for e-commerce, news, products and more
Localized RAGs
Build localized RAGs for your enterprise with support for multiple languages
Financial retrieval
Accurately retrieve unstructured financial data with understanding
Features for every developer
Structured data
All models have been trained from the ground up to response in a consistent structure on every run
Automatic scale
Serverlessly run BILLIONS of models concurrently in less than 200ms and only pay for what you use
Purpose-Built Models
Purpose-built models trained for specific tasks, delivering state-of-the-art quality and performance
Easy integration
Fully typed SDKs, clear documentation, and copy-pastable code snippets for seamless integration into any codebase
Observability
Real-time logs and analytics. Debug errors, track users, location maps, sessions, countries, IPs and 30+ data points
Secure & Private
Secure and private instance for your data. Fine grained access control on API keys.
Global first models
Multilingual
Global support for over 160+ languages across all models
Global training datasets
We collect training data from all around the world to ensure our models are as accurate no matter the locality or niche context
Distributed GPUs
90+ global GPUs to ensure the fastest inference times all the time
Smart cache
Automatic smart caching to lower cost and improve latency