Development Q&A

Framework-Related#

Q: How should the Group concept in Agent communication be understood? A: A Group is a logical communication boundary with the following characteristics:

  • Messages are broadcast only within the Group.
  • It adopts a tree structure, comprising one parent node and multiple child nodes.
  • Each node can have only one parent but multiple children.
  • Child nodes can recursively have their own sub-nodes.
  • Messages between Groups are isolated and do not interfere with one another.

    • Q: What is the mechanism for message passing between Agents? A: Message transmission between Agents adheres to the following rules:

  • Agents must belong to the same Group to communicate.
  • Messages are broadcast within the Group.
  • Messages between different Groups are isolated.
  • Message transmission is asynchronous.
  • Currently, only broadcast mode is supported.
  • Messages propagate upward only to the parent node and downward to all child nodes.

    • GAgent-Related#

    Reference for GAgent Development: Creating Your First Agent

    Q: How to handle large-scale data storage in GAgent? A: The recommended approach is:

  • Use the Event Sourcing pattern for data storage instead of directly storing it in the State.
  • For query requirements, employ the CQRS pattern:
    • The Command side handles event writing.
    • The Query side maintains a dedicated read model.
    • Event-driven synchronization ensures data consistency between read and write models.
  • Utilise storage systems for bulk data.
  • Store only essential runtime data in the State.

    • Q: How to manage event requests and responses? A: Suggested approach:

  • Subscribe to the required response event types.
  • Include a unique correlation ID in the request event.
  • Attach the same correlation ID to the response event.
  • Match requests and responses using the correlation ID.
  • Implement a timeout mechanism to handle delayed responses.

    • Q: How to return real-time responses to clients? A: For scenarios requiring real-time responses:

  • Avoid event-driven approaches.
  • Directly invoke Grain interfaces instead.
  • Rationale:
    • Grain calls provide synchronous semantics.
    • Reduces event propagation overhead.
    • Simplifies system complexity.

    • Q: How to send events to a specific Agent? A: Implement the following:

  • Add a target Agent identifier field to the event.
  • The receiver must validate if the event is intended for them.
  • Discard the event if the target identifier does not match.
  • Use Agent IDs or other unique identifiers for targeting.

    • Station-Related#

    Station Interface Documentation: Aevatar StationVI Interface Documentation

    Q: How can applications provide external web services? A: Use one of the following methods:

  • Leverage the WebHook mechanism provided by Station.
  • Refer to the WebHook section in the Station interface documentation.
  • Review the WebhookTemplate sample code: https://github.com/aevatarAI/aevatar-station/blob/dev/templates/WebhookTemplate/src/WebhookTemplate/Handler/TelegramWebhookHandlers.cs

    • Q: Does creating a GAgent interface support specifying GrainId? A: Current status:

  • The current version does not support specifying GrainId.
  • GrainId is auto-generated by the system.
  • Future versions may include this feature.
  • Ensure GrainId uniqueness if implementing custom logic.

    • Q: Can Station's Webhook invoke Grains in Silo? A: Yes, interaction is supported:

  • Webhooks can invoke Grains within Silo.
  • Ensure proper configuration of Silo connection details.

    • Edited on: 24 February 2025 07:00:59 GMT+0