GraphQL API vs GUI: Agents execution not available via API (agents.run → "agent service not configured")

[marcoantani]

Hi,
I’m integrating Moltis via the GraphQL API and trying to understand the difference in behavior between the GUI and the API when it comes to Agents execution.

Setup

• Moltis running on: https://localhost:34871/graphql
• GraphQL accessible via GraphiQL interface and HTTP requests with Authorization: Bearer <api_key>
• GUI works correctly and can execute agents
• GraphQL API works correctly for general queries (status, config, identity, etc.)

Expected behavior
Based on the GraphQL schema, the following mutation exists:

mutation {
  agents {
    run(input: JSON!)
  }
}

Also available in schema:

• runWait(input: JSON!)
• updateIdentity
• updateSoul

So I expect agent execution to be available via API.

Actual behavior
Using the GUI I created and configured two agents, so considering also the default agent I have 3 agents, active and working perfectly via GUI chat.

When executing:
{ agents { identity { name emoji theme } } }
Returns:
{ "data": { "agents": { "identity": { "name": "moltis", "emoji": null, "theme": null } } } }
No custom agents configured through the GUI are visible via the API.
In the GUI, the same agent shows customized properties (such as emoji and theme), but these values are not reflected in the API response.

Another inconsistence: agents.list returns an empty array

Question
Is the Agents runtime supposed to be available via GraphQL API in all builds?

If yes:

• What configuration or service needs to be enabled?
• Is there a missing backend service registration (e.g. agent service not mounted in the runtime)?

If not:

• Is agent execution intentionally only available via GUI/internal runtime?
• What is the recommended way to trigger agents programmatically from external applications?

I am trying to understand how agent execution is intended to work in Moltis when accessed via the GraphQL API compared to the GUI.

Specifically, I am attempting to trigger an agent through the API using structured input and have it execute a task involving MCP filesystem access and return a structured response.

The goal is to confirm whether this execution flow is officially supported via the API or if it is only available through the GUI/internal runtime.

Any clarification on the correct architecture or required configuration would be highly appreciated.