On-Premise & Self-Hosted

On-premise AI data analysis for sensitive spreadsheets

Deploy RowSpeak with Docker inside your own network, connect DeepSeek, GLM, or another open-source model, and turn finance and operations files into reviewable answers and reports without sending data to a public AI service.

Book Demo Get the Deployment Docs

Docker self-deployment • Fully air-gapped option • Deployment docs sent after a short form

Laptop showing a RowSpeak spreadsheet analysis dashboard in front of on-premise server racks

Private environment

Data stays inside your approved boundary

Why Teams Choose Us

Files, prompts, and outputs stay in your network

Self-host with Docker, with deployment docs included

Run fully air-gapped, with no internet connection

Use DeepSeek, GLM, and other open-source models

From private model hosting to a usable industry workflow

A private model endpoint alone does not reconcile a workbook, explain a variance, or produce a report that finance can review and sign off.

RowSpeak Private Deployment adds the workflow layer around the approved model.

It turns messy exports and multi-sheet workbooks into a repeatable industry workflow: ingest the file, run deterministic checks, apply a finance or operations template, and return evidence-backed output inside your environment.

With RowSpeak Private Deployment, your team can:

Keep files, prompts, and outputs inside your approved environment
Route simple tasks to an efficient open-weight model when quality and hardware checks pass
Use industry templates for reconciliation, variance review, and recurring reports
Keep calculations deterministic and show the evidence behind each answer
Deliver output your team can review, correct, share, and use

The RowSpeak Solution

Approved data boundary
Swappable model layer
Industry workflow templates
Reviewable outputs

Why teams choose a private workflow agent

Keep sensitive data where it belongs

Process spreadsheets and reporting workflows inside your own environment instead of relying on a shared hosted setup.

Test the model against the workflow

Benchmark candidate open-weight MoE or approved API models on your real files, then route each task to the model that meets quality, latency, and cost requirements.

Fit existing IT and security requirements

Deploy RowSpeak in a way that aligns with internal review, access control, and infrastructure standards.

Package the work by industry

Start with one painful workflow—such as monthly reconciliation or budget-versus-actual review—and make its inputs, checks, prompts, and outputs repeatable.

Measure an outcome, not a demo

Use a proof of concept to compare review time, correction rate, answer quality, and cost per completed workflow before a wider rollout.

Self-hosted with Docker: deploy in your own environment

RowSpeak ships as Docker containers. Your IT team can install it with our deployment docs, or work with us remotely. Pick the boundary that matches your data policy.

On-Premise

Run RowSpeak on servers in your own data center. Files, prompts, model calls, and logs stay on infrastructure you control.

Fully Air-Gapped

Run RowSpeak and the model in an isolated network with no internet connection, for teams whose policy rules out any outbound traffic.

Private Cloud

Deploy in your own cloud account or private VPC when your approved environment is cloud-based.

Private LLM support: DeepSeek, GLM, and other open-source models

RowSpeak runs on top of an open-source model you host yourself, so no prompt or file content goes to a public model API. Test candidate models on your own files, then choose the one that fits your quality and hardware budget.

DeepSeek Open Source

GLM Open Source

Qwen Open Source

Kimi Open Source

gpt-oss Open Source

Llama Open Source

Mistral Open Source

Other open-source LLMs Open Source

Model quality varies slightly between models and versions. Final results depend on your files and hardware, so we recommend testing on your own data first.

Why this matters:

Benchmark quality on representative files before choosing a model

Route different workflows to different models when needed

Compare total workflow cost, not only token price

Keep the model layer replaceable as open weights evolve

On-premise vs cloud AI data analysis

Both use the same RowSpeak workflow. The difference is where your files and the model run.

Question	RowSpeak Cloud	RowSpeak On-Premise
Where files are processed	RowSpeak cloud, in encrypted storage	Your own servers or private cloud
Which model answers	Models managed by RowSpeak	An open-source model you host, such as DeepSeek or GLM
Internet connection	Required	Optional. Fully air-gapped is supported
Setup	Sign up and upload a file	Docker deployment by your IT team, with our docs
Best for	Getting started fast with non-restricted files	Finance, HR, and regulated data that cannot leave your network

Hardware requirements for self-hosted DeepSeek and GLM

Most of the hardware cost comes from the model, not from RowSpeak. The figures below are the minimum GPU setups published by the model makers and their recommended inference frameworks.

Model	Size	Published GPU setup	Source
GLM-4.7-Flash	30B (3B active)	1× H100 (BF16)	Z.ai GLM README
GLM-4.5-Air	106B (12B active)	2× H100 (FP8) / 4× H100 (BF16)	Z.ai GLM README
GLM-4.5 / 4.6 / 4.7	355B (32B active)	8× H100 (FP8) / 16× H100 (BF16)	Z.ai GLM README
DeepSeek V3 / R1	671B (37B active)	1 node of 8× H200, or 2 nodes of 8× H100 / H20	SGLang DeepSeek guide

Figures come from official vendor documentation and assume FP8-capable GPUs such as NVIDIA H100/H200. Longer context windows and more concurrent users need more GPUs. Both model families also document support for Huawei Ascend and AMD accelerators. We confirm sizing for your workload in the deployment docs.

What your team can put into production

The deliverable is a controlled workflow that turns real business files into evidence-backed analysis and report-ready output.

Data Boundary

Model Choice

Workflow Fit

Reviewability

Your team can:

Start from real files

Use the spreadsheets and exports your team already reviews

Run repeatable checks

Check columns, totals, dates, duplicates, and missing values before analysis

Apply a workflow template

Reconcile, compare, explain, or summarize with a known business process

Show evidence

Keep source rows, formulas, assumptions, and corrections visible to reviewers

Export the decision pack

Share a report, chart, or dashboard after a human review

The opportunity is a repeatable result: less manual work, clear evidence, and a workflow the team can trust.

Built for teams with a repeatable, sensitive workflow

RowSpeak Private Deployment is a strong fit for organizations that:

Cannot upload sensitive files to a shared SaaS environment

Have a repeated, document-heavy workflow with a clear owner

Need more control over infrastructure and model selection

Need finance, operations, or legal reviewers to check the output

Want to measure business results before expanding the deployment

Common teams include:

Finance

Operations

RevOps

Strategy

Business systems

Internal analytics teams

How deployment works: from Docker image to first analysis

1. Request the deployment docs

Fill in the short form. We send the deployment guide, architecture reference, and sizing notes to your inbox.

2. Prepare the server and model

Choose an open-source model such as DeepSeek or GLM and provision GPUs that match its published requirements.

3. Run RowSpeak with Docker

Start the RowSpeak containers inside your network, on-premise, in a private cloud, or fully air-gapped.

4. Test on your own files

Run a repeated workflow, such as a reconciliation or monthly report, on representative files and check the output.

5. Review and expand

Have reviewers sign off on the results, then roll out to more users and workflows.

Book Demo

Get the self-hosted deployment docs

Tell us a little about your environment and we will email the deployment guide for running RowSpeak with Docker.

RowSpeak Self-Hosted Deployment Pack

PDF • Updated regularly

Docker deployment guide

Reference architectures, including air-gapped

Hardware and model sizing guidance

Deployment and review checklist

Built for technical review and business adoption

RowSpeak Private Deployment is designed to help both technical and business stakeholders move forward with confidence.

Technical teams can review:

Deployment architecture
Data flow boundaries
Environment ownership
Model routing choices
Infrastructure requirements

Business teams can evaluate:

Reporting workflow fit
Speed to usable output
Reviewability of results
Internal rollout potential

The goal is a reviewable workflow that technical, security, and domain teams can evaluate together.

Frequently Asked Questions

Is RowSpeak self-hosted?

Yes. RowSpeak can be self-hosted with Docker on your own servers or in your private cloud. Your IT team can deploy it using our deployment docs, which we send after you fill in a short form.

Can RowSpeak run fully offline (air-gapped)?

Yes. RowSpeak and the model can run in an isolated network with no internet connection.

Which open-source models are supported?

RowSpeak supports popular open-source models, including DeepSeek, GLM, Qwen, Kimi, gpt-oss, Llama, and Mistral. Results can vary slightly between models, so we recommend testing a few on your own files before choosing one for production.

What hardware do we need?

It depends mainly on the model. For example, GLM-4.5-Air runs on 2× NVIDIA H100 in FP8, and full DeepSeek V3 runs on a single 8× H200 node, according to the vendors' documentation. See the hardware table on this page. We confirm sizing for your workload in the deployment docs.

Does any data leave our network?

No. In an on-premise or air-gapped deployment, files, prompts, model calls, outputs, and logs stay inside the environment you run.

How is this different from running a local LLM?

A local LLM gives you a chat endpoint. RowSpeak adds the spreadsheet workflow around it: reading Excel, CSV, and PDF files, running calculations, building charts and reports, and showing the steps so a reviewer can check the result.

Can we get help with deployment?

Yes. Our team can work with your IT and security staff remotely to review the architecture and rollout steps.

Learn more about Private Deployment

Data Security & Compliance Technical Architecture Finance Team Use Cases POC Process Guide Performance Benchmarks Model Selection Guide Extended FAQ

See how a private workflow agent fits your environment

Evaluate a private RowSpeak workflow agent against your infrastructure, security requirements, model options, and first industry use case.