Amgix

Amgix is an open-source hybrid search system designed to simplify the process of implementing advanced search capabilities over real-world data. It addresses the common challenge of integrating multiple components—such as embedding services, brokers, vector databases, and fusion logic—by providing a unified REST API that manages ingestion, embedding, and hybrid retrieval within a single system. This approach eliminates the need for developers to write glue code or manually orchestrate separate services for search functionality.

The platform offers several features tailored to both developers and operators. For developers, Amgix streamlines the process of adding search to applications by allowing documents to be posted directly to the system, which then handles tasks like queueing, deduplication, distributed locking, retries, and embedding generation internally. Hybrid search is supported through server-side fusion, enabling the use of dense vectors, sparse models such as SPLADE, and keyword tokens within a single collection, with fusion logic executed on the server.

Operators benefit from built-in monitoring tools, including a dashboard and metrics endpoints that provide real-time insights and support integration with JSON and Prometheus metrics. The system is flexible in deployment, running as a single container via Amgix One for ease of setup, and scaling out to distributed architectures as needed. Encoder workers are designed for autonomous operation, routing work based on capacity and demand without requiring manual model-to-machine assignments. Amgix is compatible with existing databases such as PostgreSQL and MariaDB for vector storage and asynchronous ingestion, and can also scale to use Qdrant for larger deployments, all while maintaining a consistent API.

For end users, Amgix introduces WMTR (Weighted Multilevel Token Representation), which combines multiple lexical views—such as surface form, normalized tokens, and character-level patterns—into a single sparse representation. This design improves retrieval performance for data with unique identifiers like part numbers or SKUs, which are often mishandled by traditional tokenizers. The full search pipeline is engineered to deliver low-latency results, blending keyword precision with semantic relevance in every query. Amgix is available as open-source software and can be deployed using containerization tools such as Docker.