Separate interface and execution
Study the workbench’s layers when deciding which part of a workflow should collect context and which part should execute the task.
Study ZCode’s workbench and agent-runtime design to improve your own workflows. Use its interface separation and public source as references for organizing task context, execution feedback, and human review around work in Floatboat.
ZCode is Z.ai’s coding-agent harness and development workbench. The official repository describes desktop, browser, and terminal experiences, and includes client code, backend services, shared UI, and the agent CLI and runtime. It is broader than an editor plugin or a single task instruction.
Its value as a workflow reference is the relationship between the interface, backend services, and agent execution. A Floatboat user can study those boundaries to decide where context should enter, how a task’s progress should be presented, and when the user should review a proposed action. Actual reuse requires a separately verified integration.
The public source uses Apache-2.0 and the repository also provides a project notice covering additional operational and third-party matters. Source availability does not settle model access, hosted service terms, or every application behavior. Review the official notice and release documentation for the version you choose.
Understand what the project demonstrates, then decide which mechanism addresses a real gap in your own workflow.
Study the workbench’s layers when deciding which part of a workflow should collect context and which part should execute the task.
Use the application structure as a reference for presenting agent activity alongside the context a person needs to judge it.
Read client, backend, CLI, and runtime source together when evaluating a workflow pattern rather than copying an isolated interface.
Use these scenarios as design references for work organized in Floatboat. Any software integration needs its own compatibility and permission checks.
Define how source material, current activity, and expected deliverables should stay connected during a longer task.
Plan what evidence and context a user needs before accepting an agent’s proposed change or external action.
Study where state and execution belong when the same work can be viewed through different interfaces.
Start with a task and a small testable change, whether that means better instructions, a new tool connection, or a runtime component.
Separate business context, user interaction, execution tools, and the result that must be reviewed.
Use the project notice and source to assess what is local, what is remote, and which service receives task information.
Test one task from input to a reviewable result before expanding the workflow across more tools or interfaces.
Compare independent agents through their interfaces, extension models, and approach to carrying work forward.