Why does GEAK v4 depend on Claude Code, unlike v3.2.2 where it acted as a standalone Agent?

[Quasimodo0808]

Question

In version v3.2.2, GEAK was treated as an Agent like Claude Code.
Why does v4 need to depend on Claude Code to run?
Is there any design purpose behind this?

[zihaoanllm]

Thanks for the question. This is an intentional design change in GEAK v4.

Claude Code provides the general agent runtime and handles tool execution and environment interaction, while GEAK focuses on GPU optimization through deterministic workflows, specialized agents, profiling, benchmarking, and validation.

This separation allows GEAK to support more structured and reliable multi-agent workflows without maintaining a separate agent runtime.

[Quasimodo0808]

Thanks for your reply, this design leverages CC's harness engineering