Superlinked Inference Engine (SIE)

Superlinked builds SIE, the Superlinked Inference Engine, which serves small open models for everyday AI tasks. According to its privacy policy the controller is Superlinked, Inc. in San Francisco; the site does not state a founding year. The product is positioned for search, RAG and document workflows that do not need a frontier model, and it integrates with vector databases such as Chroma, LanceDB, Qdrant and Weaviate.

SIE is open source under Apache 2.0 and runs in three ways: SIE Serverless (pay-per-call API), SIE Local (an open-source server for a laptop or GPU VM) and Native Inference (a cluster in the customer's cloud with Superlinked engineers). The site lists 142 models covering semantic search, reranking, hybrid search, ColBERT, PDF-to-Markdown, OCR, PII redaction and speech to text.

• Search and RAG models: semantic search, reranking, hybrid search, ColBERT and image search.
• Document processing: PDF to Markdown, field extraction and OCR.
• Safety: guardrails and PII redaction.
• SIE Serverless: pay-per-call hosted API with $20 launch credit for new accounts.
• SIE Local: Apache-2.0 server for a laptop or GPU VM.
• Native Inference: SIE Cluster in the customer's own cloud.

• A Vietnamese enterprise adding semantic search and reranking over internal documents.
• A bank or insurer extracting fields from scanned forms and redacting PII.
• An e-commerce platform running image and hybrid product search through one inference API.

• Usage-based Serverless pricing means no GPUs to manage for small teams.
• Open-source Apache 2.0 server allows offline or self-hosted use.
• One API for embeddings, reranking, OCR and extraction reduces the number of services.
• SOC 2 Type 2 and a choice of US or EU data residency on Serverless support compliance review.

Teams are replacing large general models with smaller task-specific open models to cut cost and latency in search and document pipelines, and want inference they can run in their own cloud. Vietnamese companies handling Vietnamese documents and customer data may prefer self-hosted or regional inference for cost and data control.

Superlinked, Inc., San Francisco (per privacy policy); founding year not stated. SOC 2 Type 2 listed. Open-source components under Apache 2.0. Integration partners include Chroma, LanceDB, Qdrant and Weaviate.