Useful Checks

This section explains most common errors and the checks to perform for troubleshooting

"Model not found" error

• Check if the agent id in request matches the GenAI model record id

• Check if the GenAI model is configured to be allowed for the specific runner (or all runners)

• Check if the runner LangChain handler has a "domain" specificiation that doesn't match the model's domain configuration

• Check if the "channel" in request URL path matches the specific runner

Agent not remembering chat history

• Check if the agent has "memory" parameter configured for an existing state manager on the runner

• Check if the agent memory state manager is a shared state (e.g. not in-memory), if there are multiple replicas running for the runner

Only agents with memory utilize instructions & tools, so it is recommended to havethe "memory" parameter defined except for lightweight chat use cases.

Agent not executing tool as expected

• Ask the agent to provide the request body and the response details (such as errors) it is receiving from calling the tool

• If the tool is a saga flow:

  • Try to access saga as an API with the same request and review response

  • Check to make sure that the saga is allowed to run on the runner of the agent

  • Check to make sure that the user making the request has rights to execute the saga (if it is restricted)

  • Check to make sure that the saga returns a "message" field in its response

  • Check to make sure that the saga input schema is mapped accurately for the flow logic

• If the tool is a state processor:

  • Check to make sure that the target state is available on the runner of the agent

  • Check to make sure that the user making the request has rights to access the state (if it is restricted)

• If the tool is a system processor:

  • Check to make sure that the target system is available on the runner of the agent and has access details configured correctly

  • Create a simple saga to test system accessibility through an API