Hyland open-sources OpenArch for enterprise AI agents

By Olivia Hart

Hyland has put OpenArch on GitHub under an Apache 2.0 license, giving developers an open-source content index for connecting enterprise repositories to AI services and agents without stripping away the context around each file.

The project extends the content-management vendor into agentic AI. OpenArch connects data repositories built on platforms including Alfresco and Nuxeo, which Hyland acquired in 2020 and 2021 respectively, to AI services and agents.

Hyland CEO Jitesh Ghai tapped enterprise content-management veteran John Newton to lead the project after bringing him out of retirement. Newton founded open-source Alfresco in 2005 and left after Hyland acquired it in 2020. Before that, he co-founded Documentum, which OpenText later acquired. He rejoined Hyland in 2025 as chief innovation strategist.

Alan Pelz-Sharpe, founder of independent research firm Deep Analysis, said Newton's ECM and open-source experience should help Hyland's customers move faster into the agentic AI era.

Why the index matters

OpenArch catalogs metadata, permissions, business rules and other signals surrounding enterprise content:

• Metadata describing the file and its place in the business.
• Permissions governing who or what can access it.
• Business rules attached to the content.
• Other signals that can shape retrieval and compliance.

Together, those records help govern what an AI agent can retrieve and support compliance with corporate and regulatory policies. The goal is not merely to find a document; it is to preserve the history, security controls and metadata that travel with the original.

"This is a good thing," Pelz-Sharpe said. "Usually, AI just strips the data out of the file and dumps the file, so they've got no history, no security, no permissions, no metadata. [AI tries] to invent its own metadata."

OpenArch is not itself an AI harness. Pelz-Sharpe said it can form part of a broader harness strategy that gives agents governed access to content.

Why Hyland opened the stack

Hyland's free connectors can extend beyond the company's customer ecosystem, making the open-source release a distribution strategy as well as a product decision. Pelz-Sharpe said Hyland understands open source as a way to unlock bigger projects when the time comes to connect.

His assessment was blunt: nobody wants to buy a connector, and nobody wants to issue an RFP for one. Open source, he said, offers a clever route to traction outside Hyland's core user base.

What changes for enterprise deployments?

Newton said OpenArch was designed to run on-premises or in hyperscaler environments, depending on the user's requirements. Developers can deploy it themselves, connect selected systems, inspect how permissions are handled, and keep content and AI inside their own infrastructure when required.

That addresses a common enterprise problem: content remains scattered across repositories, each carrying its own permissions and rules. A file might sit in a legacy ECM system, an operational repository or a compliance system. It might also live in SharePoint, Salesforce or ServiceNow.

Newton said most enterprises cannot identify where a given piece of information lives. Each location can add another layer of access controls for an agent to understand and respect.

"An agent shouldn't have to care where that content lives; where it lives may inform the agent, but it shouldn't be the agent's problem," Newton said. "OpenArch is this layer underneath. It indexes content across the repositories, keeps all the permissions and metadata with it, and gives an agent one place to retrieve the originals."

For an enterprise AI team, the practical change is a single governed retrieval layer rather than a separate agent integration for every repository. Teams retain control over deployment and can expose original content without discarding its permissions or metadata.

OpenArch does not replace those repositories or supply the agent itself. It gives those systems a common index, and that is where Hyland is betting the next stage of agentic AI will begin.