Hyland Open-Sources OpenArch to Wire Enterprise Content Into AI

Hyland released OpenArch, an open-source content hub, on November 10, 2026, publishing it on GitHub under the Apache 2.0 license.

The software indexes unstructured enterprise content across repositories so AI models, services and agents can retrieve it. It connects sources including Alfresco, Nuxeo and Microsoft SharePoint to the AI stack a team chooses, and carries metadata and access permissions into search and retrieval so results reflect what each user is already allowed to see.

Hyland positions OpenArch for developers and AI architects who work with content spread across repositories that have different structures, permissions and rules.

What does OpenArch actually index?

OpenArch continuously builds a unified database containing content chunks, vectors used for semantic search, enterprise metadata and permissions pulled from connected repositories. The AI-ready representation lives in OpenArch. Original content and its governance stay in the source systems.

Existing repositories remain authoritative. Hyland contrasts that architecture with a data lake, arguing no second system of record gets created.

The company draws another distinction: point-to-point integrations and API gateways connect one system to another, one connection at a time. OpenArch instead maintains a common governed index across connected repositories, giving AI applications a consolidated view without replacing the underlying systems.

How does the permission model work?

Permissions are not rebuilt in a separate AI environment. On its OpenArch product page, Hyland says the software resolves user permissions dynamically at query time by calling source-repository APIs directly.

Because OpenArch does not cache or duplicate security policies, access changes made in a source system apply immediately to AI queries. That removes the separate authorization model that could drift from the authoritative system.

The repository documentation lists Alfresco, Nuxeo and Microsoft SharePoint as examples of unstructured-data sources, describes integration with data-curation pipelines for storing metadata, chunks and vectors, and presents indexing over replication as a way to limit storage-cost growth. Its API-first design exposes that data to a team's chosen large language models and AI services while respecting permissions during retrieval.

What can developers search and deploy?

The search layer supports keyword queries, advanced queries with filters, and semantic, or vector, search. A functional demonstration linked from the repository indexes content from Alfresco and Nuxeo, giving developers a working integration example.

Teams can self-manage OpenArch and choose where it runs, which AI services it uses and where their data resides. A Docker-based quick start is available. Docker Compose supports deployment on cloud or on-premises servers, including behind an organization's own firewall.

Hyland's product documentation says OpenArch works with AI services, large language models and development frameworks that support standard APIs. Teams can therefore swap AI components as their architecture evolves.

Does it depend on Hyland's paid platform?

OpenArch does not require Hyland's Content Innovation Cloud to function, though the two can be integrated. The open-source foundation also lets developers inspect, extend, modify and operate the access layer directly instead of depending on proprietary code they cannot examine.

“With OpenArch, we're not asking developers to migrate content, rebuild governance, or lock themselves into any one vendor's stack to put AI to work,” said John Newton, Hyland's chief innovation strategist.

OpenArch follows Alfresco and Nuxeo in Hyland's open-source history. Both earlier products are now part of the company.

What it means for enterprise buyers

For development teams, the practical result is a reusable content and permissions layer. AI applications can be built against one hub while repositories continue to own the content and enforce access. That reduces the need to recreate repository connections and authorization logic as AI use expands.

The test will be whether enterprises trust a query-time permission call against live repository APIs with real latency and uptime exposure. Hyland has put the code and the license in public. Buyers now have to decide if a shared index is cheaper than rebuilding governance for every AI project.