I am trying to use the AI Agent LLM Connector to invoke an LLM that is deployed inside a private environment (private VM / network).
Because the LLM is hosted in a private machine and cannot be accessed from the public internet, it is not possible for Camunda 8 SaaS to directly invoke this LLM using connectors.
To address this, my understanding is that:
-
The Camunda Connector Runtime should be deployed on a private Ubuntu VM, within the same network as the LLM.
-
The runtime will operate in hybrid mode, making only outbound calls to Camunda 8 SaaS to poll for tasks.
-
The connector runtime will then invoke the LLM locally within the private network.
This satisfies the no‑inbound‑connectivity constraint while still allowing Camunda SaaS to orchestrate the process.
Question 1 – Task Definition Type in Element Template
From the documentation, I understand that the connector type is defined using:
zeebe:taskDefinition:type
and that we need to modify this value to a custom connector type, for example:
JSON
“value”: “io.camunda.agenticai:aiagent:1”
Show more lines
However, in the provided element template JSON, I do not see the zeebe:taskDefinition:type field at all in the below template json connectors/connectors/agentic-ai/element-templates/hybrid/agenticai-aiagent-outbound-connector-hybrid.json at b360ed61439926ad213702d413e0caefc03f60c8 · camunda/connectors · GitHub
-
Am I missing something?
-
Is this field expected to be added manually when creating a custom element template?
-
Or is there a different mechanism by which the taskDefinition type is derived for AI Agent connectors?
Question 2 – Publishing and Using the Custom Connector Template
My current understanding is:
-
I need to copy the element template JSON given here connectors/connectors/agentic-ai/element-templates/hybrid/agenticai-aiagent-outbound-connector-hybrid.json at b360ed61439926ad213702d413e0caefc03f60c8 · camunda/connectors · GitHub
-
The task definition type is: Do i need to change ??
io.camunda.agenticai:aiagent:1 -
Publish this template to the Camunda modeler
-
Use the same task definition type when running the Connector Runtime, for example:
JSON
docker run --rm --name=HybridConnectorRuntime \
-e CAMUNDA_CLIENT_MODE=saas \
-e CAMUNDA_CLIENT_CLOUD_CLUSTERID=“<CLUSTER_ID>” \
-e CAMUNDA_CLIENT_CLOUD_REGION=“” \
-e CAMUNDA_CLIENT_AUTH_CLIENTID=“<CLIENT_ID>” \
-e CAMUNDA_CLIENT_AUTH_CLIENTSECRET=“<CLIENT_SECRET>” \
-e CONNECTOR_AGENTICAI_AIAGENT_TYPE=“io.camunda.agenticai:aiagent:1” \
camunda/connectors-bundle:
``
Show more lines
Can someone please confirm:
-
Whether
CONNECTOR_AGENTICAI_AIAGENT_TYPEis the correct environment variable name -
Whether
io.camunda.agenticai:aiagent:1is the correct connector type value for the AI Agent connector
Question 3 – Connector Type Naming Convention Confusion
In the given examples, I see the following environment variable being used:
Shell
-e CONNECTOR_HTTP_REST_TYPE=“io.camunda:http-json:local”
``
Show more lines
However, in the official element template JSON:
https://github.com/camunda/connectors/blob/7a0b1a143028511c72ffeca718b7c2f94c5a7832/connectors/http/rest/element-templates/hybrid/http-json-connector-hybrid.json
the connector type is defined as:
"io.camunda:http-json:1"
This is confusing to me:
-
Why does the example use
:localbut the template uses:1? -
Is
:localjust an alias used by the runtime? -
For custom / AI Agent connectors, should we always use the
:1version style?