shivvr
Ephemeral semantic embedding and cognitive agent service
Last verified:
What is shivvr?
Shivvr is an ephemeral semantic embedding and cognitive agent service that provides chunking, embedding, and hybrid search capabilities as a managed cloud service. It cuts text into pieces, embeds them using ONNX-based models, stores them in sled, and enables semantic search to find them again. The service is designed for AI agents that need memory and recall capabilities without requiring users to download and run 5-10GB embedding models locally.
Key features include automatic text chunking with configurable sizes, semantic embedding generation, vector storage and indexing, hybrid search combining semantic and keyword matching, vector inversion capabilities, in-memory corpus building for one-shot queries, and a RESTful HTTP API running on port 8080 in Docker. The service is open source under BSD-3-Clause license and written in Rust for performance.
Shivvr is for developers building RAG (Retrieval-Augmented Generation) systems, AI agents needing semantic memory, teams wanting to avoid vendor lock-in to major cloud embedding providers, developers who want embedding capability without local model overhead, and anyone building semantic search, question answering, document retrieval, or knowledge base applications. It runs as a managed cloud service at nuts.services but is also self-hostable via Docker.
shivvr pricing
Pricing model: Freemium
Shivvr is open source under BSD-3-Clause license and can be run locally via Docker at no cost. As a managed cloud service at nuts.services, specific pricing tiers are not publicly disclosed on the available documentation. The service is self-hostable when you're ready to bring it in-house, eliminating ongoing cloud costs. No free tier limits or paid plan details are explicitly published in the available documentation.
shivvr pros
- Open source with BSD-3-Clause license for full control
- Written in Rust for high performance and low memory usage
- Runs in Docker on port 8080 for easy deployment
- Includes vector inversion capabilities for advanced queries
- Supports in-memory corpus building for one-shot queries
- Provides hybrid search combining semantic and keyword matching
- Automatic text chunking with configurable chunk sizes
- No need to download 5-10GB embedding models locally
- RESTful HTTP API for simple integration
- Self-hostable when ready to bring in-house
- Part of N.U.T.S. microservices network with GrubCrawler
- Light and fast - doesn't burden your machine with dead weight
- Supports ONNX-based embedding models for flexibility
- Built for sovereign agentic infrastructure without cloud dependency
- Designed specifically for AI agent memory and recall needs
shivvr cons
- Managed cloud service may have usage limits not publicly disclosed
- Self-hosting requires Docker Desktop and 8080 port availability
- May require tuning for optimal chunk size for specific use cases
- Documentation is still work in progress per DeepBlue Dynamics
- eventId ephemeral nature means data may decay or be lost
- Not as established as major cloud embedding providers like OpenAI
- Vector inversion adds complexity that may not be needed for basic use
- Rust-based means fewer community contributors than Python alternatives
- May lack enterprise features like multi-tenant support or advanced monitoring
Frequently asked questions about shivvr
What is Shivvr and what does it do?
Shivvr is an ephemeral semantic embedding and cognitive agent service that performs chunking, embedding, and hybrid search. It cuts text into pieces, embeds them using ONNX models, stores them in sled, and enables you to find them again through semantic search. It's designed as a memory service for AI agents.
How do I deploy Shivvr?
Shivvr runs in Docker on port 8080. You can pull it and run it anywhere as a self-hosted service. It requires Docker Desktop and exposes an HTTP API. Alternatively, you can use the managed cloud service at nuts.services without self-hosting.
What embedding models does Shivvr support?
Shivvr uses ONNX-based embedding models, which provide flexibility for different model choices. The ONNX framework allows running various transformer-based embedding models without requiring specific hardware accelerators.
Can I use Shivvr for RAG systems?
Yes, Shivvr is specifically designed for RAG (Retrieval-Augmented Generation) systems. It provides the chunking and embedding infrastructure needed for the second stage of RAG pipelines, converting text into embeddings and storing them in a vector database for retrieval.
What is the difference between managed cloud and self-hosted Shivvr?
The managed cloud service runs at nuts.services and requires no setup on your part. Self-hosted Shivvr runs in your own Docker container on port 8080, giving you full control and eliminating ongoing cloud costs, but requires you to manage the infrastructure.
What is vector inversion in Shivvr?
Vector inversion is an advanced capability included in Shivvr that allows for specialized query types beyond standard semantic search. It enables reverse vector operations that can be useful for certain complex retrieval scenarios in AI agent applications.
How does Shivvr handle text chunking?
Shivvr automatically cuts text into pieces (chunks) before embedding. The chunking is configurable, and you can build an in-memory corpus for one-shot queries. Proper chunking ensures embeddings focus on local context rather than diluted global document semantics.
Is Shivvr open source?
Yes, Shivvr is open source under the BSD-3-Clause license. The source code is available on GitHub at DeepBlueDynamics/shivvr, and you can inspect, modify, and self-host it without licensing restrictions.
What programming language is Shivvr built with?
Shivvr is built in Rust, which provides high performance, low memory usage, and safety guarantees. The Rust implementation combined with ONNX for embeddings and sled for storage makes it lightweight and fast.
Who should use Shivvr?
Shivvr is for developers building AI agents that need semantic memory, teams building RAG systems, developers wanting to avoid vendor lock-in to major cloud providers, and anyone who needs embedding capability without downloading 5-10GB models locally. It's part of DeepBlue Dynamics' sovereign agentic infrastructure suite.