How nxtStudio
Works

A modular, plugin-driven platform built for enterprise AI workloads. Explore the architecture that powers secure document intelligence across federal and state agencies.

Request a Demo Explore Capabilities

Architecture Layers

Production Plugins

20+

AI Models Available

Tenant Hierarchy Levels

Platform + Plugins = Infinite Flexibility

A layered architecture where each concern is isolated, composable, and independently extensible. Add new capabilities without touching the core.

User Interface

Three-Panel WritingFLR DashboardChat InterfaceAdmin Console

Plugin Layer

AI WritingDoc AnalysisFLRChatbot+ New

Platform Layer

Auth / SSOMulti-TenantAI ProvidersRAG / VectorRolesAudit LogsDoc StoragePersonas

Data Layer

PostgreSQL + pgvectorFile StorageEmbeddings

Each layer communicates through well-defined APIs. Plugins can be developed, tested, and deployed independently — without modifying the platform core.

Enterprise Separation Built In, Not Bolted On

A 6-level organizational hierarchy ensures complete data separation while supporting flexible team structures across agencies and teams.

Organizational Hierarchy

Instance

Dedicated deployment

Tenant

Contract / agency boundary

Organization

Agency or team

Domain

Subject area or program

Workspace

Team collaboration space

Project

Individual work unit

Variable Org Level

Not every tenant needs all six levels. Configure the hierarchy to match your organizational structure.

Cascading Permissions

Roles granted at a parent scope automatically cascade to children. Override at any level for fine-grained control.

Scoped Roles

Users can belong to multiple workspaces and organizations simultaneously with different roles in each.

Dedicated Instances

For maximum separation, deploy a completely separate instance per contract with independent infrastructure.

Choose Your AI. At Any Prompt.

nxtStudio's adapter architecture means all providers implement a common interface. Swap models per-prompt without changing a single line of application code.

BaseProviderAdapter

Abstract base class

sendMessage()streamMessage()discoverModels()testConnection()estimateTokens()

OllamaAdapter

Ollama (Local)

Air-gapped, local inference. Data never leaves your network.

AnthropicAdapter

Anthropic Claude

Nuanced reasoning with industry-leading long context windows.

OpenAIAdapter

OpenAI GPT

Structured output and function calling for data extraction.

BedrockAdapter

AWS Bedrock

Enterprise-managed, VPC-deployed, AWS-native integration.

The security advantage: Analyze sensitive documents with a local Ollama model, then switch to Claude for non-sensitive synthesis — all in the same session, same interface.

From Upload to Insight in Minutes

The RAG pipeline transforms raw documents into searchable, AI-ready knowledge with full configurability at each stage.

Upload

PDF, DOCX, TXT, and more

Extract

Text & structure extraction

Chunk

Intelligent segmentation

Embed

Vector representations

Index

pgvector similarity search

Generate

Grounded AI responses

Upload

Drag-and-drop or API upload with automatic format detection

Extract

Apache Tika-powered extraction preserving headings, tables, and metadata

Chunk

Configurable chunk size and overlap with boundary-aware splitting

Embed

Generate dense embeddings using your chosen model provider

Index

HNSW indexing for sub-second semantic retrieval across millions of chunks

Generate

Context-aware generation with citation tracking and source attribution

Configurable Parameters

Each stage of the pipeline is tunable per-project

Chunk Size

1,000 tokens

200 - 4,000 tokens

Chunk Overlap

200 tokens

0 - 1,000 tokens

Top-K Results

1 - 20

Similarity Threshold

0.7

0.0 - 1.0

Generation Mode

RAG

RAG, Factual, Augmented, Creative

Context Window

Model max

Configurable per request

Parameter

Default

Range

Before & After nxtStudio

See the concrete impact across common enterprise document workflows.

Task

Before

After

Document Analysis

Manually read thousands of pages

Semantic search across entire corpus in seconds

Content Writing

Write from memory, hope citations are right

AI drafts grounded in source docs with tracked citations

Literature Review

Months of manual screening and extraction

AI-powered screening with dual-reviewer workflows

AI Model Selection

Locked into one vendor, one model

Choose the best model per-prompt, switch freely

Security & Isolation

Shared infrastructure, trust the vendor

Dedicated instances, multi-tenant scoping, full audit logs

Onboarding New Teams

Weeks of configuration and custom development

Create workspace, assign roles, upload documents, go

How nxtStudio Compares

Generic AI chatbots lack grounding. Single-purpose tools lack flexibility. nxtStudio delivers both.

Capability

Generic AI Tools

Single-Purpose Tools

nxtStudio

Document Grounding (RAG)

Basic or none

Single-method

Full pipeline, configurable

AI Model Choice

One vendor locked

Any model, any prompt

Multi-Tenant Separation

Shared infrastructure

6-level hierarchy, dedicated instances

Systematic Review (FLR)

Not available

Limited methodology

PRISMA, OHAT, Cochrane, JBI

Citation Tracking

None or unreliable

Partial

Source-level attribution per claim

Self-Hosted / Air-Gapped

Cloud only

Varies

Full support with Ollama

Plugin Extensibility

Closed system

Add plugins without core changes

Audit Logging

Minimal

Comprehensive, every action

See the Platform in Action

Our demos run on the production platform — not slides. Tell us about your workflow and we'll show you exactly how nxtStudio handles it.