agentic-ui

What problem does it solve?

Users struggle to trust and control AI actions when decisions happen outside chat, especially during multi-step workflows, streaming output, tool usage, and failures.

Core Features & Use Cases

• Intent Preview: Show users exactly what the AI will do before execution with Approve/Edit/Cancel and affected-item details.
• Autonomy Control: Let users choose Manual, Supervised, or Autonomous behavior per feature to match their trust level.
• Task Decomposition UI: Break goals into visible steps with per-step status, progress, and approve/reject controls.
• Streaming States + Stop: Provide token-streaming UX with differentiated “thinking/searching/generating” states and an always-available stop control.
• Tool Call Cards + Citations: Display tool usage transparency (inputs/outputs/durations) and inline sources to support verification.
• Confidence Indicators + Error Recovery: Communicate uncertainty with confidence rings and offer concrete recovery paths instead of dead-end failures.

Quick Start

Use the agentic-ui patterns to design an AI feature where users can preview the intended action, select autonomy level, watch streaming progress, review tool/citation details, and recover gracefully if something fails.