openclaw - 💡(How to fix) Fix Feature: Configure dedicated LLM model for Dreaming subsystem [1 comments, 2 participants]

TL;DR

Implement a configuration option to specify a dedicated model for the Dreaming subsystem, allowing for cost reduction and performance improvement.

Guidance

• Add a dreaming.model configuration key to the memory-core plugin, as proposed in the issue, to enable specification of a dedicated model for Dreaming tasks.
• Update the documentation to reflect the new configuration option and its benefits, including cost reduction and performance improvement.
• Test the new configuration option with a smaller model, such as ollama/qwen2.5:7b, to verify its effectiveness in offloading cloud calls and improving performance.
• Consider adding a fallback mechanism to ensure backwards compatibility, where the agent's primary model is used if the dedicated Dreaming model is not specified.

Example

plugins:
  entries:
    memory-core:
      config:
        dreaming:
          model: ollama/qwen2.5:7b

This example demonstrates how to configure a dedicated model for the Dreaming subsystem using the proposed dreaming.model key.

Notes

The proposed solution assumes that the smaller model is capable of handling the structured and low-complexity tasks of the Dreaming subsystem. Further testing and evaluation may be necessary to determine the optimal model for this use case.

Recommendation

Apply the proposed workaround by adding the dreaming.model configuration key and updating the documentation, as this will allow for cost reduction and performance improvement while maintaining backwards compatibility.