Private LLM for a Law Firm: The Questions Firms Ask Before They Build One

Seven questions midsize firms bring to a private LLM build, answered by a Microsoft partner working on the platform for 20 years.

New York, NY, Sept. 24, 2026 (GLOBE NEWSWIRE) -- A private LLM for a law firm runs as a commercial model inside the firm's Microsoft cloud subscription, and eSoftware Associates (ESW), a 100% U.S.-based Microsoft partner on the platform for 20 years across more than 1,500 organizations, has formalized the answers it gives midsize firms scoping that build. ESW has completed 200+ migrations into Microsoft 365 GCC and GCC High environments with zero data loss.

"Security in these builds comes from the deployment. A firm can pick a careful vendor and still send its matter files to a service sitting outside its network controls," said Russell Kommer, founder and CEO of eSoftware Associates.

Question: What is a private LLM for a law firm, exactly?

Answer: It’s a commercial language model running inside the firm's cloud subscription, where prompts and documents stay in the firm's tenant and aren’t retained for training. A law firm adopting one chooses a deployment pattern, and the model itself still comes from a commercial provider. ESW provisions it in the firm's Azure subscription and grounds it on the firm's SharePoint libraries and matter records so answers come back with citations an attorney can open.

Question: Do we have to train our own model, or is there a shortcut?

Answer: Retrieval is the standard approach, in which the model reads the firm's documents at the moment of the query instead of memorizing them in advance, so a new precedent is available as soon as it’s filed rather than after the next training run. Training a model on a firm's matter history is a separate research project carrying its maintenance obligation. ESW has delivered thousands of custom Microsoft solutions for more than 1,500 clients, and scopes the retrieval build against a firm's matter volume before a firm budgets for a custom model.

Question: Does a private LLM run on our own servers or in the cloud?

Answer: In a private deployment, it runs in the firm's cloud subscription. ESW provisions the model inside the firm's Azure tenant, which puts the network boundary, logging and retention settings under the firm's administrators while Microsoft carries the hardware. On-premises hosting moves GPU purchase, capacity planning and patching onto the firm, while the Azure route already carries regulated-environment controls. ESW's Azure management practice supports environments governed by HIPAA, FINRA and GDPR requirements, and a law firm's confidentiality obligations run on those same controls.

Question: Can the AI see matters an attorney is walled off from?

Answer: Not if the wall is enforced as a permission. Retrieval runs as the person who typed the query, so the model can only assemble an answer from documents that person is already authorized to open, and sensitivity labels applied to privileged material hold the same way. That protection is only as strong as the permissions underneath it, so a wall that exists only in a policy memo does not constrain a model. ESW audits SharePoint and file-share access before connecting the model, then verifies each wall by running test queries as a walled user instead of reading the configuration back.

"A configuration tells a firm what it set up, and running the query as a walled user tells the firm what the wall actually does. Those two answers don’t always match," Kommer said.

Question: Will it know our precedent and our templates?

Answer: Yes, once it’s connected to them, and the answers stay only as current as the library itself. The capability comes from retrieval against the firm's document libraries, so the model reads current precedent at query time and returns citations to specific files. A retired template will be quoted back to a partner unless the firm's records mark it as superseded. ESW resolves version history, ownership and retention alongside the connection.

Question: Is our data used to train somebody else's model?

Answer: In a private Azure deployment, prompts and outputs are not used to train the underlying foundation models, and that commitment comes from Microsoft's platform terms. ESW configures the boundary around it, restricting network access so the service is reachable only from the firm's environment, keeping logging inside the firm's tenant and setting retention to match the firm's file policy, all settings a firm's administrator can open and read. With a packaged legal AI product, the same assurances come from contract terms that vary by vendor and move at renewal.

Question: Who do we hire to build this, an AI company or a Microsoft partner?

Answer: A Microsoft partner, if the firm wants to own what gets built. A private LLM build is Azure provisioning, Copilot Studio configuration, SharePoint permission remediation and content governance, none of which is model research, so scoping starts with a Microsoft 365 AI readiness assessment of the firm's existing data and permissions. ESW holds a 92% multi-project continuation rate and staffs every build with U.S.-based consultants, with no offshore handoffs. Firms already running Microsoft 365 are extending licensing they already have, so the engagement scopes as configuration and integration.

A private LLM for a law firm outlasts the model it starts with. Commercial models are replaced on the provider's schedule, and the firm's document libraries, permission model and retention rules stay in place for whatever reads them next. ESW builds Copilot and AI agents for legal workflows on that footing, with Microsoft Azure management maintaining the network boundary and the audit trail under the firm's administrators as the model inside it changes. The access work a firm does once carries into every model it runs afterward.

About eSoftware Associates (ESW)

ESW builds secure business apps, workflows, portals, and automations across Power Apps, SharePoint, Dataverse, SPFx, Copilot & AI Agents, and Azure. 100% U.S.-based since 2006.

Sarah Evans
Head of PR, Zen Media
sarah@zenmedia.com