Camunda Agentic Orchestration FAQ Questions

An AI agent is a software program that autonomously gathers data and carries out tasks using this information, independently or on behalf of another system or person. In many ways, Camunda treats AI agents like any other task, allowing us to orchestrate them flexibly. You can also build AI agents directly in Camunda, easily granting them access to tools and embedding them directly in deterministic process flows. Learn more here.

Camunda’s agentic orchestration coordinates AI agents right alongside all your other tasks, ensuring the work of the agents is transparent and reliable. These AI agents can each individually or collectively be granted access to specific tools, governed by specific rules, and are fully auditable. Learn more here.

Using an AI agent means allowing an agent to take some independent action within your process. Agentic orchestration, on the other hand, allows you to coordinate multiple AI agents, or even a single agent with access to multiple tools, so that the agent(s) work effectively. Learn more about the differences here.

Deterministic orchestration proceeds along a predetermined path and does not deviate from that other than by specifically determined rules. Dynamic orchestration allows actions to be determined as the process instance is executed by leveraging things such as AI models and runtime data.

Most business processes have traditionally been deterministic, and this has many advantages. With Camunda, it’s easy to take a deterministic process and add a dynamic element to it using our ad-hoc sub-process or connectors, enabling you to add a dynamic AI agent directly into your existing processes. Learn more about this here.

Many AI agent builders struggle to provide explainability, but Camunda creates comprehensive audit trails that capture the context and process around every AI interaction, and captures larger process metrics within Camunda Optimize. This is particularly true with agents built directly in Camunda (third-party agents may have certain data limitations). Read more about this here.

In many cases, organizations have specific policies where LLMs are approved for certain kinds of information. If your organization does not have policies that make it clear where secured data can be stored, Camunda encourages data pseudonymization to enable agentic orchestration use cases.

BPMN is a widely used standard that provides end-to-end visibility for a process that both business and technical users can understand. Camunda uses BPMN to allow you to retain this same visibility and composability with agentic processes, to avoid the “black box” of a solution that uses AI in a non-explainable way. Learn more about this here.

Camunda uses BPMN for process modeling, which gives you the ability to control the high-level logic of your AI agents with advanced BPMN patterns, including timing, escalation, humans in the loop and more. For agents built in Camunda, we also have technical guardrails that allow you to specify the maximum model calls, maximum tokens, and more. You can also easily grant individual agents access to a specific tool or tools. Essentially, there are guardrails both around the agent (in the overall process) and within the agent itself. Learn more about how Camunda provides governance and control for AI agents.

Because Camunda blends deterministic and dynamic orchestration, you can easily add in human involvement where necessary, whether it is to review a final outcome or for consultation in the midst of an ongoing AI-driven process. Learn more about how BPMN helps with this here.

Yes. Camunda is a composable platform and you can use most major LLMs with it. Camunda provides a prebuilt AI Agent connector to make this as easy as possible.

Best practices to secure against prompt injection include security by design and the use of prompt injection detection frameworks. Camunda makes it easy to add these to your process with our unique BPMN-based modeling, which means you can simply add security features as a step between your data and the Agent without changing any other parts of your process.

Camunda supports both short-term and long-term memory for AI agents. Short-term memory (for the agent itself) can be saved within the process itself, but when desired, we also support transferring this memory to a relevant document through our native document handling capabilities. Long-term memory (to be shared with others) can be shared with other humans or other AI agents using our out-of-the-box vector database connector. Learn more about this in our documentation and our Marketplace.

Our process orchestration platform manages state with our Zeebe engine, which enables us to have long-running state management for AI agents (just as you can for anything else in your business process). For example, in our AI agent connector, by default the AI agent will simply wait for a customer response as long as necessary.

AI agents can learn through short-term memory (for example, a response from a customer within a process) as well as by calling tools (which provide it with information). AI agents can also utilize long-term memory to access accumulated data. See an example here.

In the Camunda Marketplace you’ll find many examples with best-practice design patterns for agentic orchestration, which you can use directly or adapt to your own needs.

AI agents have access to everything that Camunda provides, including connectors, job workers, and more. In an upcoming alpha release we will be releasing an MCP connector as well to make using that easier, but you don’t have to use MCP. Camunda allows you to choose what works best for you.

We don’t distinguish between AI agent tools or any other job worker task. They are all tools that are executed and orchestrated by Zeebe. This enables AI agents to seamlessly integrate into existing processes and benefit from the existing horizontal scalability, resilience and security available to all Camunda processes.

An AI agent in Camunda can access any BPMN element, including launching a new workflow or calling other agents. We don’t distinguish between calling new tasks or calling a new sub-process or AI agent—Camunda orchestrates all of these just as easily.

Yes, there are. Learn more about agentic orchestration in action here, and see examples in Camunda Marketplace that you can adapt to your use case today. Of course, you can also contact us directly and request a customized demonstration.