Agent performance upgrades and improved map visualizations

In response to developer feedback, this release focuses on significant performance optimizations and component upgrades across the Maps Agentic UI (MAUI) toolkit. It includes a new agent using Server-Side Template Architecture in MAUIAgentWithTemplates (~80% faster than the previous MCP-based grounding agent), and an updated agent with Grounding Latency Optimization in agent_with_grounding (reducing latency by ~40%). It also contains enhanced 3D Map Marker & Label Rendering in the Web SDK.

What's Changed

⚡ Server-Side Template Agent (MAUIAgentWithTemplates)

• Latency Reduction: Targets end-to-end response times of P50 ~6.0s for Local Search and Directions (down from >25s–40s in dynamic synthesis baselines).
• 2-Agent Fast Routing Pipeline: Bypasses from-scratch LLM UI layout synthesis for common spatial queries by separating intent classification from parameter extraction:
  • Fast Router: Classifies incoming query intent (e.g., LOCAL_SEARCH, DIRECTIONS, TEXT_ONLY) using lightweight models (gemini-3.1-flash-lite).
  • Parameter Extractor: Extracts structured spatial parameters into validated schemas.
  • Server Merger: Programmatically merges extracted parameters into pre-compiled server layout templates (templates/local_search.json, templates/directions.json).
• Guaranteed Valid Schema: Generates 100% valid JSON component layouts for templated spatial queries, eliminating client-side stream repair retries and rendering errors.

⚡ Vertex AI Grounding Latency Optimization (~40% Latency Reduction)

• Latency Reduction: Avoids secondary LLM tool-calling turns, reducing end-to-end response latency by ~40% while preserving strict grounding accuracy.
• Single-Turn Grounded Generation: In agent_with_grounding.py, the agent resolves Google Maps place IDs directly from grounding metadata during single-pass generation.

🗺️ 3D Map Marker & Label Rendering Enhancements

• Label Collision Avoidance: Configured OPTIONAL_AND_HIDES_LOWER_PRIORITY collision behavior on 3D labels to prevent overlapping text on dense map results.
• Anchor & Route Pins: Added styled pins for anchor locations, destination points, and route origin/destination pairs.

🛠️ Demo Server Configuration

• Configurable Agent Selection: The Python demo server supports configuring the active agent using the --agent CLI option or the A2UI_DEFAULT_AGENT environment variable.