Hyland Open-Sources OpenArch to Wire Enterprise Content Into AI

Hyland released OpenArch on November 10, 2026, an open-source content hub that makes unstructured enterprise content available to AI models, services and agents — available immediately on GitHub under the Apache 2.0 license. The software targets a problem that has stalled enterprise AI adoption: content scattered across repositories with different structures, permissions and rules, none of it natively reachable by AI stacks. Hyland's answer is an indexing layer, not a migration. That distinction carries most of the story's weight.

What does OpenArch actually do?

OpenArch connects unstructured content sources — the repository lists Alfresco, Nuxeo and Microsoft SharePoint as examples — to whatever AI stack a team chooses. It continuously builds a unified database containing content chunks (fragmented text used for retrieval), vectors used for semantic search, enterprise metadata and permissions drawn from connected repositories. The original content and its governance stay in the source systems. Hyland explicitly contrasts this architecture with a data lake, because no second system of record is created. Existing repositories remain authoritative.

The company also distinguishes the standing index from point-to-point integrations and API gateways, which it describes as connecting 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. The takeaway: teams get one reusable access layer instead of a growing web of bespoke connectors.

How does it handle permissions?

This is the part enterprise architects will scrutinize hardest. Permissions are not rebuilt inside a separate AI environment. According to Hyland's OpenArch product page, the software resolves user permissions dynamically at query time by calling source-repository APIs directly. OpenArch does not cache or duplicate security policies. Because of that, access changes made in a source system apply immediately to AI queries, avoiding a separate authorization model that could drift from the authoritative system.

The claim is credible in design but unproven at scale. Query-time permission checks against live repository APIs trade correctness for latency; Hyland has not published performance figures for how that resolution behaves under load, and the announcement is silent on throughput. Buyers should verify that trade-off against their own query volumes before committing.

Metadata and access permissions also carry into search and retrieval, so results reflect what each user is already allowed to see. In plain words: if a user cannot open a document in SharePoint, the AI should not surface its contents either — at least in principle, pending independent testing.

What does the developer experience look like?

The repository details several concrete capabilities:

• Integration with data-curation pipelines for storing metadata, chunks and vectors
• Indexing over replication, presented as a way to limit storage-cost growth
• An API-first design exposing indexed data to a team's chosen large language models and AI services, respecting permissions during retrieval
• A search layer supporting keyword queries, advanced queries with filters, and semantic (vector) search
• A functional demonstration that indexes content from Alfresco and Nuxeo as an integration example

Deployment is self-managed. Teams choose where OpenArch runs, which AI services it uses and where their data resides. A Docker-based quick start is available, and 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, so teams can replace AI components as their architecture evolves. The short version: no mandatory cloud, no mandatory model provider, no mandatory stack.

Why open-source, and why now?

OpenArch does not require Hyland's Content Innovation Cloud to function, although the two can be integrated. The open-source foundation lets developers inspect, extend, modify and operate the access layer directly rather than depend on proprietary code they cannot examine. For a company whose commercial business depends on enterprise content management, giving away the connectivity layer is a calculated move: it lowers the barrier for AI teams to standardize on Hyland's plumbing, with the monetization presumably arriving further up the stack.

"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. The quote is a direct pitch at the three objections that most often kill enterprise AI pilots — data movement cost, governance duplication and vendor lock-in. Notably absent from the quote and the announcement: any benchmark claims, customer names or production deployments.

OpenArch follows Alfresco and Nuxeo in Hyland's open-source history, with both earlier products now part of the company. Newton, incidentally, is an Alfresco co-founder; the open-source lineage here is long-standing rather than a new experiment.

What is the practical payoff for development teams?

The implication, per Hyland, 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 across an organization. Each new agent or assistant connects to OpenArch rather than renegotiating access to every underlying system.

The context matters. Enterprise AI deployments routinely stall not on model quality but on the unglamorous problem of getting governed content to models without breaking compliance. Retrieval-augmented generation pipelines (systems that fetch relevant documents to ground model answers) live or die on exactly the indexing and permission layer OpenArch provides. Hyland is competing for that infrastructure slot against every major cloud vendor and a crowd of vector-database startups.

The caveats deserve equal billing. The initial source list — Alfresco, Nuxeo, SharePoint — covers Hyland's own heritage platforms plus one Microsoft staple; connectors for the wider enterprise estate remain unstated. Performance, scale limits and enterprise support terms are likewise absent from the announcement. What exists today on GitHub is a foundation, not a finished ecosystem.

Still, an Apache 2.0 license with no dependency on Hyland's commercial cloud gives teams a genuinely inspectable option, and the query-time permission model — if it performs — addresses the most common governance failure in enterprise AI retrieval. Whether OpenArch becomes standard infrastructure or a curiosity will depend on connector breadth and adoption beyond Hyland's installed base.