Natural Language Processing

# Absolute dates
"Add a meeting on March 15th at 2pm"
"Schedule dentist appointment on 3/20/2026 at 10am"

# Relative dates
"Create an event tomorrow at 9am"
"Add a meeting next Tuesday at 2pm"
"Schedule a call in 2 hours"

# Contextual times
"Book lunch tomorrow at noon"
"Set up a meeting this afternoon"
"Add a morning standup every day"

# Duration
"Schedule a 30-minute meeting with John"
"Block 2 hours for deep work tomorrow"
"Add a quick 15-minute sync"

# Simple queries
"Show me my recent emails"
"What's in my inbox?"
"List my unread messages"

# Sender-based
"Find emails from Sarah"
"Show messages from my boss"
"List emails from john@example.com"

# Subject-based
"Search for emails about the project"
"Find messages with 'invoice' in the subject"
"Show me budget-related emails"

# Time-based
"Show emails from this week"
"List messages from yesterday"
"Find emails after March 1st"

# Combined filters
"Show unread emails from Sarah with attachments"
"Find important messages from last week"

# Create
"Create a task: buy groceries"
"Add a todo: finish report"
"Remind me to call the dentist"

# With due dates
"Add task 'Submit proposal' due Friday"
"Create a todo: review PR, due tomorrow"
"Remind me to pay rent on the 1st"

# Complete
"Mark 'buy groceries' as done"
"I finished the report"
"Check off the dentist task"

# Update
"Change my task title to 'Buy groceries and supplies'"
"Update the due date to next week"
"Add notes to the project task"

"Add a meeting tomorrow at 2pm"
# Parsed: 2026-03-09T14:00:00

"Schedule an event next Tuesday"
# Parsed: 2026-03-10T09:00:00 (defaults to 9am if no time)

"Create a task this Friday"
# Parsed: 2026-03-12T00:00:00

"Add a reminder in 2 hours"
# Calculated from current time: 2026-03-08T16:30:00

"Schedule a meeting in 3 days"
# Parsed: 2026-03-11T09:00:00

"2026-03-15T14:00:00"
"2026-03-15"  # Date only, defaults to 9am

"March 15th at 2pm"
"3/15/2026 at 2:00 PM"
"March 15, 2026"

"2pm", "2:30 PM", "10:15am"

"14:00", "22:30", "09:15"

"noon" → 12:00 PM
"midnight" → 12:00 AM
"morning" → 9:00 AM
"afternoon" → 2:00 PM
"evening" → 6:00 PM

from agent.date_utils import parse_duration

parse_duration("1 hour")     # timedelta(hours=1)
parse_duration("30 minutes") # timedelta(minutes=30)
parse_duration("2h")         # timedelta(hours=2)
parse_duration("1.5 hours")  # timedelta(hours=1.5)
parse_duration("45 min")     # timedelta(minutes=45)
parse_duration("2 days")     # timedelta(days=2)

from agent.date_utils import parse_datetime

# Natural language → ISO format
date_str = "next Tuesday at 2pm"
iso_date, is_all_day = parse_datetime(date_str, timezone="UTC")
# Returns: ("2026-03-10T14:00:00", False)

# Date without time
date_str = "March 15th"
iso_date, is_all_day = parse_datetime(date_str, timezone="UTC")
# Returns: ("2026-03-15", True)

"Add a meeting"
"Create an event"
"Schedule a call"
"Set up a reminder"
"Book an appointment"

"Show me my tasks"
"List my emails"
"What's on my calendar?"
"Display my schedule"
"What are my todos?"

"Change the meeting time"
"Update the task title"
"Move the event to Friday"
"Modify the appointment"
"Reschedule the call"

"Delete the meeting"
"Remove the task"
"Cancel the appointment"
"Delete the event"
"Remove my reminder"

"Mark the task as done"
"I finished the task"
"Complete the todo"
"Check off the task"
"Done with that task"

# User: "Show me my calendar"

# Step 1: Discover relevant tools
meta_tools.discover_tools(category="calendar", action="list")
# Returns: ["list_calendar_events"]

# Step 2: Get tool schema
meta_tools.get_tool_schema(tool_name="list_calendar_events")
# Returns: Full parameter schema

# Step 3: Invoke tool
meta_tools.invoke_tool(
    tool_name="list_calendar_events",
    parameters={"max_results": 10}
)
# Returns: List of calendar events

PLAN_ACTIONS_SYSTEM_PROMPT = """
You are an AI assistant that helps users manage their calendar, email, and tasks.

Current date: {current_date}
Current time: {current_time}

You have access to three meta-tools:
1. discover_tools - Find tools by category/action
2. get_tool_schema - Get detailed schema for a tool
3. invoke_tool - Execute a tool with parameters

Use these meta-tools to fulfill the user's request.
"""

SYNTHESIZE_RESPONSE_PROMPT = """
You are summarizing the results of tool executions for the user.

Provide a natural, helpful response that:
- Confirms what was done
- Presents results clearly
- Explains any errors
- Suggests next steps if appropriate

Format lists and data nicely for readability.
"""

# From agent state
conversation_history = state.get("conversation_history", [])

# Recent messages (last 5) are included in prompts
for msg in conversation_history[-5:]:
    if msg.get("role") == "user":
        messages.append(HumanMessage(content=msg["content"]))
    elif msg.get("role") == "assistant":
        messages.append(AIMessage(content=msg["content"]))

# User: "Add a meeting with John next Tuesday at 2pm for 1 hour"

# Extracted parameters:
{
    "tool_name": "add_calendar_event",
    "parameters": {
        "title": "Meeting with John",
        "start_time": "2026-03-10T14:00:00",
        "end_time": "2026-03-10T15:00:00",
        "attendees": ["john@example.com"],  # If email known
        "timezone": "UTC"
    }
}

User: "Add a meeting with John"
Agent: "I'd be happy to schedule that. When would you like to meet with John?"

User: "Tomorrow at 2pm"
Agent: "Meeting with John scheduled for March 9th at 2:00 PM."