AI agents

Build and integrate AI agents into your end-to-end processes.

About AI agents

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.

AI agents can perform a variety of functions, including making decisions, solving problems, interacting with external environments, and taking actions.

The AI Agent connector

The AI Agent connector is the primary Camunda connector for building AI agents. It integrates an LLM with your BPMN process, enabling the agent to reason over context, select tools, and respond to users or process events.

Key capabilities include:

• LLM provider support: Connects to a range of providers, such as Anthropic, Amazon Bedrock, Google Gemini, and OpenAI.
• Tool calling: Exposes BPMN activities inside an ad-hoc sub-processAd-hoc sub-processA special type of BPMN subprocess that allows activities to be executed in any order, skipped, or repeated, without a predefined sequence. Activities are activated dynamically at runtime, either by a human or a system, rather than following a fixed flow. as tools the LLM can select.
• Memory: Short-term conversational memory enables multi-turn interactions and follow-up questions within a process instance.

See the AI Agent connector documentation for full configuration details, implementation examples, and reference.

Integrate an AI agent into your process

The recommended approach for most use cases is to use the AI Agent Sub-process implementation due to its simplified configuration and support for event sub-processes.

In this approach, you integrate the agent using an ad-hoc sub-process and the AI Agent connector in a tool feedback loop, where the agent understands the process goal and uses the available tools to complete it.

How the feedback loop works

The AI Agent connector operates in a feedback loop between the LLM and Camunda:

1. A user prompt is sent to the connector. The LLM evaluates the prompt, the system prompt, and the available tool definitions.
2. If the LLM determines that a tool call is needed, Camunda activates the corresponding BPMN activity in the ad-hoc sub-process.
3. The tool result is passed back to the LLM, which decides whether more tool calls are needed.
4. The loop continues until the LLM returns a final response, which can then be routed to the next step in the process.

Decision-making and execution are intentionally split:

• LLM decides: Which tool to call next, in what order, and with which parameters.
• Camunda orchestrates: Executes the selected BPMN activity, stores variables, applies retries and incident handling, and routes human tasks and events.

tip

Learn more in the example AI Agent Sub-process connector integration and guide to adding a tool for an AI agent.

AI agent integration features

Use the following Camunda 8 features to integrate AI agents into your processes:

Feature	Description
Ad-hoc sub-process	A special kind of embedded BPMN subprocess with an ad-hoc marker that allows a small part of your process decision-making to be handed over to a human or agent.
AI Agent connector	Enables AI agents to integrate with an LLM to provide interaction/reasoning capabilities. This connector is designed for use with an ad-hoc sub-process in a feedback loop, providing automated user interaction and tool selection.
MCP Client connector	Connect an AI agent connector to tools exposed by Model Context Protocol (MCP) servers.
Ad-hoc tools schema resolver connector	Can be used independently with other AI connectors for direct LLM interaction. Use this connector if you don't want to use the AI agent connector but still want to resolve tools for an ad-hoc sub-process or debug tool definitions.
Vector database connector	Allows embedding, storing, and retrieving LLM embeddings. Use this connector to build AI-based solutions such as context document search, long-term memory for LLMs, and agentic AI interaction.