adk-ui

Dynamic UI generation for AI agents. Enables agents to render rich user interfaces through tool calls.

Features

• 30 Component Types: Text, buttons, forms, tables, charts, modals, toasts, and more
• 13 Render Tools: High-level tools for common UI patterns, including protocol-aware screen/page emitters
• 10 Pre-built Templates: Registration, login, dashboard, settings, and more
• Bidirectional Data Flow: Forms submit data back to agents via UiEvent
• Streaming Updates: Patch components by ID with UiUpdate
• Server-side Validation: Catch malformed responses before they reach the client
• Type-Safe: Full Rust schema with TypeScript types for React client
• Protocol Interop: Emit UI payloads as a2ui, ag_ui, or MCP Apps (mcp_apps)

Quick Start

[dependencies]
adk-ui = "0.3.2"

use adk_ui::{UiToolset, UI_AGENT_PROMPT};
use adk_agent::LlmAgentBuilder;

// Add all 13 UI tools to an agent with the tested system prompt
let tools = UiToolset::all_tools();
let mut builder = LlmAgentBuilder::new("assistant")
    .model(model)
    .instruction(UI_AGENT_PROMPT);  // Tested prompt for reliable tool usage

for tool in tools {
    builder = builder.tool(tool);
}

let agent = builder.build()?;

Modules

Prompts (prompts.rs)

Tested system prompts for reliable LLM tool usage:

use adk_ui::{UI_AGENT_PROMPT, UI_AGENT_PROMPT_SHORT};

// UI_AGENT_PROMPT includes:
// - Critical rules for tool usage
// - Tool selection guide
// - Few-shot examples with JSON parameters

Templates (templates.rs)

Pre-built UI patterns:

use adk_ui::{render_template, UiTemplate, TemplateData};

let response = render_template(UiTemplate::Registration, TemplateData::default());

Templates: Registration, Login, UserProfile, Settings, ConfirmDelete, StatusDashboard, DataTable, SuccessMessage, ErrorMessage, Loading

Validation (validation.rs)

Server-side validation:

use adk_ui::{validate_ui_response, UiResponse};

let result = validate_ui_response(&ui_response);
if let Err(errors) = result {
    eprintln!("Validation errors: {:?}", errors);
}

Available Tools

Streaming Updates

Update specific components by ID without re-rendering:

use adk_ui::{UiUpdate, Component, Progress};

let update = UiUpdate::replace(
    "progress-bar",
    Component::Progress(Progress {
        id: Some("progress-bar".to_string()),
        value: 75,
        label: Some("75%".to_string()),
    }),
);

Protocol Outputs

All 13 tools support protocol-aware output selection through the protocol argument.

Supported protocol values:

• a2ui
• ag_ui
• mcp_apps

Output Matrix

Example args:

{
  "protocol": "mcp_apps",
  "mcp_apps": {
    "resource_uri": "ui://demo/surface"
  }
}

adk-ui now includes matrix coverage tests for all 13 x 3 tool/protocol combinations in adk-ui/tests/tool_protocol_matrix_tests.rs.

Migration guidance for legacy/default outputs is documented in adk-ui/docs/PROTOCOL_MIGRATION.md.

Deprecation Timeline

Legacy runtime profile adk_ui now carries explicit deprecation metadata:

• announced: 2026-02-07
• sunset target: 2026-12-31
• replacements: a2ui, ag_ui, mcp_apps

This metadata is exposed through shared capability constants (UI_PROTOCOL_CAPABILITIES) and surfaced by adk-server in /api/ui/capabilities.

Interop Adapters

adk-ui includes adapter primitives for protocol conversion from canonical surfaces:

• A2uiAdapter
• AgUiAdapter
• McpAppsAdapter

These adapters implement a shared UiProtocolAdapter trait and are used by render tools to avoid per-tool protocol conversion drift.

React Client

Install the npm package:

npm install @zavora-ai/adk-ui-react

import { Renderer } from '@zavora-ai/adk-ui-react';
import type { UiResponse, UiEvent } from '@zavora-ai/adk-ui-react';

Or use the reference implementation in examples/ui_react_client/.

Examples

Architecture

Agent ──[render_screen/render_page]──> protocol payload (`a2ui` | `ag_ui` | `mcp_apps`)
                 ↑
                 └────────── UiEvent / action feedback loop

License

Apache-2.0

Part of ADK-Rust

This crate is part of the ADK-Rust framework for building AI agents in Rust.