Core Concepts
Understand the building blocks of Neutron
Architecture Overview
Seeds
The fundamental unit of content in Neutron
A Seed represents any document or text content you upload. When you upload a PDF, text file, or raw text, Neutron creates a Seed record and processes it into searchable chunks.
Supported Types
Seed Properties
id - Unique identifier
title - Display name
type - Document type (pdf, text, etc.)
status - Processing status (pending, processing,
completed, failed)
tokenValue - Total tokens in the document
Chunks
Smaller pieces of content optimized for search
During processing, each Seed is split into Chunks of approximately 300 tokens. Each chunk gets its own embedding vector (1024 dimensions) for semantic search.
Chunking Details
- Target size: ~300 tokens
- Smart splitting at paragraphs/sentences
- Preserves code blocks and tables
- Maintains document structure
Embeddings
- Model: Jina Embeddings v4
- Dimensions: 1024
- Multimodal: Text and images
- Multilingual support
Bundles
Organize seeds into logical groups
Bundles are containers for organizing related Seeds. A seed can belong to multiple bundles, and you can filter searches by bundle.
// Create a bundle
POST /bundles
{ "name": "Research Papers", "color": "#8B5CF6" }
// Add seeds to bundle
POST /bundles/{bundleId}/seeds
{ "seedIds": ["seed-1", "seed-2"] }External Users
Multi-tenant data isolation
The externalUserId parameter provides data isolation
between your users. Each user can only access their own seeds, chunks, and bundles.
Hierarchy
Organization (your account)
└── App (project)
└── External User (your customer)
├── Seeds
├── Bundles
└── Chunks