Description
LightOn offers a powerful AI retrieval API designed to "Point Your AI at Your Documents." It provides three core endpoints: `/parse` for document structuring, `/extract` for pulling specific fields based on a defined schema, and `/search` for grounded retrieval with citations.
Read More
The `/parse` endpoint utilizes LightOnOCR-2, a state-of-the-art parsing engine that transforms scans, tables, handwriting, and multi-column layouts into structured Markdown. It supports over 20 languages natively and is built on open weights available on HuggingFace, achieving 83.2 on the OLMOCR-BENCH.
The `/extract` endpoint allows users to define a JSON schema and receive structured JSON output, enabling the extraction of specific entities like invoice numbers, lease end dates, or contract clauses. This feature is priced per page and supports asynchronous processing via webhooks.
The `/search` endpoint provides grounded retrieval with citations, leveraging a hybrid approach combining dense, sparse, and late-interaction signals. Built on the open-source ColBERT family (LateOn and NextPlaid), it ensures every search result is accompanied by the exact source passage that produced it. This feature is priced per query with low latency.
LightOn is built for agents, handling raw PDFs, garbled tables, and off-domain queries with robust retrieval capabilities. It emphasizes trust through grounded results and auditable design. The platform is LLM-agnostic, allowing users to bring their own models, and supports VPC, on-prem, or air-gapped deployments for enhanced security and compliance.
Key use cases include counterparty risk analysis, multi-jurisdiction contract review, and regulatory perimeter mapping, where instant discovery of knowledge and surfacing insights are critical. The platform offers features like single sign-on, role-based access, and GDPR/SOC 2 compliance. A free tier is available for developers to test all endpoints in the console.
- CategoryVector Databases & Retrieval
- Pricing ModelPer Page / Per Query
- DeploymentCloud, VPC, On-Premise, Air-gapped
- PlatformsWeb
LightOn AI's Core Features
Document Parsing with OCR
Structured Data Extraction
Grounded Retrieval with Citations
Multi-language Support (20+ languages)
Hybrid Search (Dense, Sparse, Late-Interaction)
LLM-Agnostic Architecture
Open-source Models (ColBERT family)
Production-ready API
Free Tier Available
EU-hosted Infrastructure
Role-based Access Control
Workspace and ACL Scoping at Chunk Level
Data Synchronization from External Sources
How to use LightOn AI?
Access Console: Try endpoints in the playground console.
Configure API: Obtain an API key for Console and Enterprise.
Parse Documents: Use the /parse endpoint for structured Markdown output.
Extract Fields: Utilize the /extract endpoint with a defined JSON schema.
Perform Search: Employ the /search endpoint for grounded retrieval with citations.
Integrate API: Copy code snippets into your project for application integration.
Deploy Enterprise: Discuss VPC, on-prem, or air-gapped deployment options.
LightOn AI's Use Cases
- Document Analysis
- Information Retrieval
- Agent Development
- Contract Review
- Risk Assessment
- Regulatory Compliance
- Knowledge Management
- Enterprise Search
