This bridges the gap when your AI agent needs actual human input mid-workflow. It's voice-first, meaning users can speak answers instead of typing them, which gets surfaced through answer links, MCP tools, or standard HTTP APIs depending on your setup. Reach for this when you're building agents that can't make assumptions and need to ask clarifying questions, get approvals, or collect information only a human has. The streamable HTTP transport means you can integrate it without heavy SDK dependencies. Think form submissions, confirmation dialogs, or any decision point where the AI legitimately needs to hand off control and wait for a person to weigh in before continuing.