October 2026
Agent Mesh is now generally available
Agent Mesh is Hyland's governed platform for building and running agentic business solutions. It brings agents, business processes, rules, content, and human judgment together, allowing organizations to automate content-driven work while keeping control over how decisions are made and work is executed.
With Agent Mesh, Solution Builders can create solutions where agents interpret information, make context-aware decisions, coordinate work, and involve people when human judgment is required.
The initial GA release introduces built-in agents for document understanding, task execution, and orchestration, together with capabilities for building, running, monitoring, and governing agentic solutions.
Key capabilities available with Agent Mesh GA
Orchestrate work with the Orchestrator Agent
The Orchestrator Agent coordinates work across an Agent Mesh solution.
It evaluates incoming work against configured scenarios and determines what should happen next. Depending on the configuration, it can route work to the appropriate scenario or coordinate multiple steps toward a defined business goal.
This helps Solution Builders create adaptive solutions where the next action can depend on the context and results of previous work.
Work with documents using the Document Agent
The Document Agent helps Solution Builders use document understanding and reasoning directly within their solutions.
It can work with document content to perform tasks such as summarization, comparison, interpretation, and other document-driven activities based on a defined goal.
Perform flexible tasks with Task Agents
Task Agents provide a flexible way to perform AI-driven activities within an Agent Mesh solution.
Solution Builders can configure their instructions, inputs, tools, and expected outcomes to support different business tasks and use cases.
New capabilities in this release
Choose between Standard and Advanced orchestration
The Orchestrator Agent supports two orchestration modes, allowing Solution Builders to choose the level of coordination that fits their use case.
Standard is designed for routing. The Orchestrator matches incoming work to a scenario, starts the corresponding process or agent, and completes its work without interpreting the result.
Advanced is designed for end-to-end orchestration. The Orchestrator can use the results of completed scenarios to decide what should happen next, start follow-up scenarios, and continue until the orchestration reaches its goal.
With Advanced orchestration, coordination logic that would otherwise need to be modeled explicitly can be handled by the Orchestrator itself. This enables more adaptive, multi-step business processes where the next action depends on the results of previous work.
Define a business goal for the Orchestrator Agent
You can define a Business goal for an Orchestrator Agent to describe the business outcome the agent should achieve.
The goal guides the agent's decisions and helps it select the right actions and scenarios to reach the expected result.
Set execution limits for Advanced Orchestrator Agents
You can set execution limits for Advanced Orchestrator Agents to control how many times scenarios can run during an orchestration.
You can define:
- the total number of scenario runs allowed per orchestration, including repeated runs;
- the maximum number of runs allowed for individual scenarios.
These limits provide more control over complex orchestrations and help prevent scenarios from running more times than intended.
Process and analyze data with agents
Task Agents and Document Agents can use the Process and analyze data built-in tool to analyze, clean, summarize, and transform data from files.
The tool can also create new files based on the results of the analysis.
Return generated files from agents
Task Agents and Document Agents using Process and analyze data can return generated files as content that can be used in subsequent steps.
For Task Agents, Contents is available as a type in the Outcomes configuration.
For Document Agents, generated content is exposed as an additional output on the Agentic Task.
Introducing Hyland Control Tower: one place to see whether your agents are running, failing, performing well, and approved
Control Tower is Hyland's platform for monitoring and governing the agents running in your environment. Until now, teams had no single view of which agents exist, whether they were failing, or whether anyone had approved them for production. Control Tower brings that together in one place. Access is granted by assigning roles to user groups, the same way as other Hyland modules.
Dashboard: a system-wide view of agent health and activity Previously, problems with agents were found late, often through user complaints. The Dashboard is your landing page, with key metrics across your agents in one view. You can filter by time range and hide agents that are no longer available.
Agent Library: a searchable inventory of every agent in your environment Agents created through the Agent Builder API, Studio Modeler, and Knowledge Discovery / ECE (RAG and GraphRAG) now appear in one list, so you can see what exists and avoid duplicates. Search by name or description, sort by name, type, last modified, or recent activity, and copy an agent's full ID with one click, which is useful when an error message gives you an ID and you need to know which agent it belongs to. Select an agent to see its run history.
Platform Monitoring: see show-stopper errors as they happen Errors that stop an agent from completing its job used to surface through support tickets. System Health now shows these failures so you can act before users report them.
Agent Runs: every run, one click from root cause Troubleshooting a failed agent used to mean reading raw logs. Each agent now has a run list showing status, latency, and a summary, with filters by result and search across summaries. Open a run to see the input, the final output, and a latency breakdown showing where time was spent. If the agent called another agent as a tool, that run is linked. Run history is stored independently of tracing sampling. Run Details requires observability tracing to be enabled on the agent.
Passport: nothing runs in production unless it is approved Before Passport, nothing stopped an unreviewed agent from running in production. Agents now need an approved Passport to run. Submit an agent for review, then approve or reject it, with a note required on rejection. The approval is checked each time the agent is invoked, and changing the agent's configuration invalidates it, so the agent that runs is the agent that was approved. This allows to immediately stop misbehaving agents. The option can be enabled per environment.
Online Evaluations (BETA): score the quality of live agent runs An agent can finish without errors and still give a poor answer. Online Evaluations scores live runs so you can catch problems that never raise an error. View a run's evaluation results from its Run Details page. This is a beta feature, available only when enabled on the agent. Behavior and UI may change before general availability.