Basic AI Agent Connector
Camunda

Basic AI Agent Connector
Camunda

Download What_to_do_.zip file

Download What_s_the_weather_like_.zip file

Out-Of-The-Box Agentic AI from Camunda 8.8 ☛

AI Agent Connector is in fact an “ad hoc subprocess” including different activities. AI Agent Connector is a customizable AI via native facilities offered from Camunda 8.8 (e.g., fromAi FEEL function). AI Agent Connector subcontracts LLMs (e.g., “Claude Sonnet”) to AI/LLM providers (e.g., “Amazon Bedrock”)

In the spirit of “Case Management”, to-be-performed activities are decided by AI (i.e., AI agent ⇒ autonomy); Camunda 8.8 orchestration engine traces what does the AI agent: what executed activities? How many passes for what computed data? Etc.

Using Ollama for Agentic AI within Camunda 8 Run ☛

Ollama (as local AI/LLM provider) is installed in conjunction with Camunda 8 Run

macOS

curl 'http://localhost:11434/api/tags'

Windows

$result = (Invoke-WebRequest -Uri http://localhost:11434/api/tags -Method GET).Content; echo $result
$x = @{"model"="qwen3"; "prompt"="Coucou, ça va ?";}; echo $x
Invoke-WebRequest -Uri http://localhost:11434/api/generate -Method POST -Body ($x|ConvertTo-Json) -ContentType "application/json"
# OpenAI API compatible: 
$y = (@{"model"="qwen3"; "messages"=@(@{"role"="user"; "content"="Coucou, ça va ?"});}|ConvertTo-Json); echo $y
Invoke-WebRequest -Uri http://localhost:11434/v1/chat/completions -Method POST -Body $y -ContentType "application/json" -ConnectionTimeoutSeconds 300

Get available Ollama LLMs

Set up AI Agent Connector based on Ollama

Set up LLM

Execution (process)

Execution (form)

Data exchange with LLM ☛

Autonomous decision making is realized by the *SELECTION* of “tools”^* (i.e., activities as parts of the Ad hoc subprocess being “the” agent)

Agentic AI core functioning within Camunda is the fact that decision-making and orchestration are split

The necessity of exchanging data with the LLM relies on the fromAI FEEL function (not part of FEEL as standard). fromAI is in charge of pushing data to tools based on the toolCall predefined local var.

toolCallResult predefined var. (tool level) and toolCallResults predefined var. (array of id. of effectively run activities, process level) are in charge of pulling data from tools

^*Camunda Agentic AI support is based on Model Context Protocol -MCP- and as such compliant with OpenAI and Anthropic

“What's the weather like?” tool as Rest Outbound Connector

Execution (process)

Execution (form)