Datasaur Launches Forge, an AI Native Service for Private, Enterprise AI

Organizations in regulated industries are facing a growing problem in enterprise AI adoption: they cannot send sensitive data to a third-party model vendor.

Today, Datasaur is launching Forge, an AI Native Service, a new practice that embeds Datasaur engineers directly inside financial, healthcare, legal, and government organizations to design, build, and operate AI systems that run entirely within the customer’s own servers.

The practice is privacy-first by construction and model-agnostic by design, allowing organizations to deploy any frontier or open-weight model they choose on top of Datasaur’s proprietary agent harness.

The launch follows recent announcements from OpenAI and Anthropic, both of which introduced enterprise services arms designed to deploy their own proprietary models into customer operations. Datasaur’s AI Native Services is built around a different premise: regulated enterprises generally cannot send their data to a third-party model endpoint, and instead require AI systems that operate where their data already lives, under their existing controls.

“Data sovereignty is the table stakes the rest of the industry is still pretending it isn’t,” said Ivan Lee, Founder and CEO of Datasaur. “Procurement at a regulated enterprise is not going to send fifty million records through someone else’s API. They want an agent that lives inside their VPC, runs on whichever model fits the task, and stays theirs at the end of the contract. That is the service we are operationalizing.”

What AI Native Services Includes

AI Native Services is organized around three core commitments.

1. Deployments Run Inside the Customer Environment

Every deployment runs entirely inside the customer’s own cloud or on-premises environment. No customer data is sent to Datasaur or to any model vendor for inference, fine-tuning, or evaluation.

2. Deployments Are Model-Agnostic

Customers can run frontier APIs, open-weight models such as Google’s Gemma, OpenAI OSS, DeepSeek, or other fine-tuned SLMs hosted on their own GPUs, and swap between them as the frontier evolves.

Datasaur’s view is that the foundation model is a commoditizing, swappable input. The orchestration layer is the durable asset.

3. Customers Own the System

At the end of the engagement, the customer owns the operational system. Embeddings, system prompts, retrieval configuration, evaluation benchmarks, training data, and the agent harness all transfer to the customer.

Only Datasaur’s internal data engine — the platform that generates those artifacts — remains proprietary.

AI as Infrastructure, Not Just a Tool

Underneath these commitments is a broader view of what enterprise AI becomes inside regulated organizations.

Datasaur is building agents that operate as infrastructure rather than standalone tools. Tool adoption is limited by the percentage of employees willing to change workflow behavior. Infrastructure adoption is determined centrally and operates against every relevant record by default.

Existing Deployments Across Regulated Industries

Datasaur already operates this model with customers including a top-three GSIB, multiple federal agencies, Am Law 100 firms, and other Fortune 1000 organizations.

Across these deployments, AI agents handle work ranging from PII redaction at the scale of hundreds of millions of records, to legacy system automation against federal compliance deadlines, to legal document review under strict data residency requirements.

The practice will be staffed by Datasaur’s existing solutions engineering and AI delivery teams. Engagements typically begin with a paid scoping phase, followed by a production deployment in 30 to 60 days.

“The consulting arms OpenAI and Anthropic just launched are not neutral parties,” Lee said. “Their economics depend on pushing more inference through their own model APIs. That is the rent side of the buy-versus-rent decision every enterprise is about to make. Datasaur sits on the buy side. We install AI systems the customer owns and runs inside their own infrastructure, on whichever model serves the work, and our business margins are not based on collecting a percentage of every transaction five years from now.”

About Datasaur

Founded in 2019, Datasaur builds private, model-agnostic AI agents that deploy inside the customer’s own cloud.

The company’s clients include Qualtrics, Stanford University, the Federal Bureau of Investigation, and organizations across legal, finance, healthcare, and government sectors.

Datasaur is backed by Initialized Capital and angel investors including Greg Brockman and Calvin French-Owen.