Agent Ui

What it does

This skill, Agent Ui, dynamically generates user interfaces based on inference results from AI agents. It allows for interactive data exploration and decision support by providing a runtime environment with built-in tool lifecycle management (pending, progress, approval, results) and human-in-the-loop approval flows. The UI is driven by declarative JSON responses from the agent, enabling flexible widget creation and real-time token streaming.

When to use it

• When you need a user interface for interacting with an AI agent's output in a structured way.
• For applications requiring human review or approval of agent actions.
• To build interfaces that display data generated by the agent, such as forms or visualizations.
• In scenarios where real-time updates and streaming responses are important.
• When you want to incorporate client-side tools into your agent workflows.

Key capabilities

• Runtime Included: No backend logic is required for the UI component.
• Tool Lifecycle Management: Tracks tool status (pending, progress, approval, results).
• Human-in-the-Loop Approval Flows: Built-in support for human review and approval of agent actions.
• Declarative JSON UI: Generates UIs from agent responses in JSON format.
• Streaming: Provides real-time token streaming.
• Client-Side Tools: Supports tools that run directly within the browser.

Example prompts

These are not prompts to the skill, but configurations for an agent using the skill:

• proxyUrl = "/api/inference/proxy" (required to connect)
• agentConfig.core_app.ref = 'openrouter/claude-haiku-45@0fkg6xwb' (specifies a core AI model)
• agentConfig.system_prompt = 'You are helpful.' (sets the agent's persona)

Tips & gotchas

• API Proxy Required: This skill requires an API proxy route setup as described in the documentation (/api/inference/proxy).
• Environment Variable: Ensure you have set the INFERENCE_API_KEY environment variable.
• Client-Side Tools: To utilize client-side tools, use the createScopedTools function and pass it to the agent's configuration.