TUI Mode

Human includes a full-screen terminal user interface (TUI) for interactive agent sessions. It provides split panes, streaming output, tool approval prompts, and multi-session tabs.

Enabling TUI mode

human agent --tui

TUI mode requires the build flag HU_ENABLE_TUI. If TUI support is not compiled in, the command reports that TUI is unavailable.

Features

| Feature | Description | | ------------------ | ------------------------------------------------ | | Split panes | Input area and output/response area | | Streaming output | Token-by-token display as the model responds | | Tool execution | Tool calls shown in a tool log with status | | Tool approval | Prompt to allow/deny medium-risk tools (y/n) | | Multi-session tabs | Switch between conversations with Ctrl+T | | Input history | Navigate previous inputs with arrow keys | | Observer bridge | TUI observer records tool events for live status |

Build requirement

The TUI is implemented with termbox2 and is guarded by HU_ENABLE_TUI. It also requires HU_GATEWAY_POSIX (POSIX platform). On Windows or when the flag is off, --tui falls back to an error message.

Layout

• Title bar: Provider and model
• Output area: Agent responses, possibly streaming
• Tool log: List of tool calls with duration and success/failure
• Status bar: Approval prompt or last tool status
• Input area: User message input

Tool approval

When the agent requests a tool that requires approval, the TUI shows a prompt such as:

Allow <tool_name>? [y/n] args: <args>

Type y to allow or n to deny. The agent continues with the result.