RowSpeak

RowSpeak Private Deployment: Technical Architecture Overview

This document is written for IT architects, infrastructure engineers, and security reviewers evaluating RowSpeak Private Deployment. It covers system components, deployment patterns, hardware requirements, and integration considerations.

System Architecture

RowSpeak Private Deployment consists of four layers that run entirely within your environment.

RowSpeak Private Deployment — System Architecture Layer 1 — Client Web Browser Desktop PWA REST API Client Layer 2 — Application Server API Gateway Auth · Rate Limit Spreadsheet Engine Parse · Transform Output Generator Charts · Reports Audit Logger Access · Actions Layer 3 — AI / Model Layer Model Router Select by policy Open-Source Models DeepSeek · Qwen · Kimi Closed-Source (opt.) GPT · Claude · Gemini Embedding Engine Local vector index Layer 4 — Data Layer File Storage Relational DB Vector DB Audit Logs

Supported Deployment Environments

Environment Supported Notes
Bare metal Linux server ✅ Recommended for maximum performance
Docker (single host) ✅ Fastest to get running
Docker Compose ✅ Standard multi-service setup
Kubernetes ✅ For high-availability and scaling
Private cloud (AWS VPC) ✅ Deploy in your own account
Private cloud (Azure VNet) ✅ Deploy in your own subscription
Private cloud (GCP VPC) ✅ Deploy in your own project
Air-gapped (no internet) ✅ Requires pre-bundled deployment package
Windows Server ⚠️ Via Docker Desktop or WSL2

Hardware Sizing Guide

The right hardware depends on your team size, concurrency needs, and whether you run open-source models locally.

Minimum (up to 20 users, light usage)

Component Specification
CPU 8-core x86_64
RAM 32 GB
GPU NVIDIA GPU with 16 GB VRAM (for local model)
Storage 500 GB SSD
OS Ubuntu 22.04 LTS or RHEL 8+

Standard (up to 100 users, regular usage)

Component Specification
CPU 16-core x86_64
RAM 64 GB
GPU NVIDIA A10 or RTX 4090 (24 GB VRAM)
Storage 1 TB NVMe SSD
OS Ubuntu 22.04 LTS

Enterprise (100+ users, high concurrency)

Component Specification
CPU 32-core x86_64 (2 nodes)
RAM 128 GB per node
GPU NVIDIA A100 or H100 (80 GB VRAM)
Storage 2 TB NVMe SSD + shared NAS
Network 10 Gbps internal
OS Ubuntu 22.04 LTS with Kubernetes

If you use closed-source models via API (GPT, Claude, Gemini), GPU requirements are significantly reduced since inference runs externally on your API key.

Network Requirements

Inbound

  • Port 443 (HTTPS) from internal users to RowSpeak app server
  • Port 80 (HTTP, redirect to HTTPS) — optional

Internal (between components)

  • App server to model server: configurable port (default 8080)
  • App server to database: default DB ports (5432 for Postgres, etc.)
  • App server to vector DB: configurable

Outbound (internet)

  • Not required for open-source model deployments
  • Required only if using closed-source model APIs (GPT, Claude, Gemini) — outbound to respective API endpoints only
  • Required for initial model weight download during setup (can be done on a separate machine and transferred)

Supported Models

Open-Source (run locally, no internet required)

Model VRAM Required Best For
DeepSeek-V2 16–24 GB General analysis, Chinese language
Qwen2.5 16 GB Multilingual, structured data
Kimi (Moonshot) 16 GB Long context, document analysis

Closed-Source (via your own API key)

Model Provider Notes
GPT-4o OpenAI Your API key, your account
Claude 3.5 Anthropic Your API key, your account
Gemini 1.5 Pro Google Your API key, your account

You can mix models across workflows — for example, use an open-source model for sensitive financial data and a closed-source model for non-sensitive reporting tasks.

Authentication and Access Control

RowSpeak Private Deployment supports:

  • Local user accounts — managed within RowSpeak
  • SSO integration — SAML 2.0 and OIDC (contact us for configuration guide)
  • LDAP / Active Directory — available on Enterprise tier
  • Role-based access control — Admin, Editor, Viewer roles
  • Workspace isolation — separate workspaces per team or department

Audit Logging

All user actions are logged locally:

  • File uploads and deletions
  • AI queries and responses
  • Report generation events
  • Login and logout events
  • Admin configuration changes

Logs are stored in your environment and are not transmitted to RowSpeak. Log format is JSON, compatible with standard SIEM tools (Splunk, Elastic, etc.).

Update and Maintenance

RowSpeak releases updates as versioned Docker images or deployment packages.

  • Updates are applied by your IT team on your schedule
  • No automatic updates without your action
  • Release notes are provided with each version
  • Rollback is supported by reverting to the previous image version

Getting the Full Technical Reference

The RowSpeak Deployment Pack includes:

  • Reference deployment architectures (PDF)
  • Hardware sizing worksheet
  • Model compatibility matrix
  • Deployment checklist
  • Network diagram templates

Request the Deployment Pack →

For a live technical walkthrough, book a demo and we will review your specific environment together.