Proactive iOS Digital Assistant
Inspired by upcoming AI features in iOS 27, this challenge involves building a prototype of a proactive digital assistant that leverages GPT-5 for advanced reasoning and LangGraph for complex, stateful workflows. The assistant will utilize the A2A Protocol for seamless agent-to-agent communication, enabling specialized agents to collaborate and provide personalized, context-aware assistance across simulated iOS applications and services. Participants will design a graph-based multi-agent system where different 'service agents' (e.g., Calendar Agent, Messaging Agent, Smart Home Agent) interact using A2A Protocol, orchestrated by a central 'Proactive Assistant Agent'. This system should anticipate user needs, suggest actions, and execute tasks across simulated device functionalities, demonstrating a cutting-edge approach to on-device AI integration and user experience.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a proactive iOS-style digital assistant using GPT-5, LangGraph for graph-based workflows, and A2A Protocol for inter-agent communication across simulated device services.
How work is evaluated
The assistant will be evaluated on its ability to understand proactive user scenarios, accurately delegate tasks to specialized agents via A2A Protocol, and correctly execute commands through simulated iOS APIs, demonstrating proactive and context-aware assistance.
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Design a graph-based multi-agent system using LangGraph for a proactive digital assistant.
Orchestrate agent teams with OpenAI Swarm for dynamic task decomposition and execution.
Implement A2A Protocol for secure and efficient communication between specialized service agents.
Leverage GPT-5 for advanced natural language understanding and proactive decision-making.
Integrate simulated iOS device APIs for calendar, messaging, and smart home control.
Reference links and supporting material
Sample data for 2 tasks
Challenge setup
The assistant will be evaluated on its ability to understand proactive user scenarios, accurately delegate tasks to specialized agents via A2A Protocol, and correctly execute commands through simulated iOS APIs, demon...
Challenge input
{ 'user_context': { 'calendar_events': ['Meeting at 3 PM today'], 'location': 'home', 'current_time': '2:30 PM', 'message_alerts': [] }, 'user_pref...
Agent execution
No agent tool is listed for this challenge.
Evaluated output
{ 'proactive_suggestion': 'Traffic is heavy, leaving now for your 3 PM meeting is recommended.', 'executed_actions': ['Send notification'] }
- Assistant must generate a relevant and accurate proactive suggestion for the given context.
- Delegated tasks must be successfully handled by the correct specialized agent, and simulated API calls must...
- Contextual Relevance Score (LLM Score) target: 4
- Python execution harness
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
No tool records are attached.
Evaluation contract
- Assistant must generate a relevant and accurate proactive suggestion for the given context.
- Delegated tasks must be successfully handled by the correct specialized agent, and simulated API calls must succeed.
Metrics
- Contextual Relevance Score (LLM Score) · Target 4
- Response Latency (ms) · Target 1000
Workflow tasks
- Task 1
Input: { 'user_context': { 'calendar_events': ['Meeting at 3 PM today'], 'location': 'home', 'current_time': '2:30 PM', 'message_alerts': [] }, 'user_preferences': { 'travel_time_alert': true } }
Output: { 'proactive_suggestion': 'Traffic is heavy, leaving now for your 3 PM meeting is recommended.', 'executed_actions': ['Send notification'] }
Execution harness: Python
- Task 2
Input: { 'request': 'Schedule a call with John Doe for tomorrow at 10 AM', 'contact_info': { 'John Doe': '123-456-7890' } }
Output: { 'delegation_status': 'success', 'receiving_agent': 'Calendar Agent', 'tool_output': { 'calendar_event_id': 'evt123', 'status': 'scheduled' } }
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.