Skip to main content
AI Voice Agents maintain conversation context across multiple turns, enabling natural multi-turn conversations.

​
What is Context?

Context is the conversation history and information that your agent remembers:
  • Conversation History - All messages in the current session
  • Tool Results - Results from function calls
  • User Preferences - Information about the user
  • Session State - Current conversation state

​
Multi-Turn Conversations

Agents automatically maintain context across conversation turns: 
# Turn 1
User: "What's the weather in London?"
Agent: "The weather in London is cloudy, 15°C."

# Turn 2 (agent remembers London from previous turn)
User: "What about tomorrow?"
Agent: "Tomorrow in London will be sunny, 18°C."

# Turn 3 (agent remembers the conversation)
User: "And the day after?"
Agent: "The day after tomorrow in London will be partly cloudy, 16°C."

​
How Context Works

​
Session-Based Context

Each WebSocket connection creates a new session with its own context:
  • Session ID - Unique identifier for each conversation
  • Message History - All messages in the session
  • Tool Call History - All tool calls and results
  • User Metadata - Optional user information

​
Context Management

The agent automatically manages context:
  1. Stores messages - All user messages and agent responses
  2. Includes tool results - Tool execution results are added to context
  3. Maintains history - Full conversation history is available to the LLM
  4. Manages session - Each session has isolated context

​
Context in Tool Calls

Tool results are automatically included in context: 
# User asks a question
User: "What's the weather in London?"

# Agent calls tool
Agent calls: get_weather(location="London")
Tool returns: "Weather in London: cloudy, 15°C"

# Tool result is added to context
# Agent uses context to respond
Agent: "The weather in London is cloudy, 15°C."

# In next turn, agent remembers the tool call
User: "What about tomorrow?"
# Agent can use previous context to understand "tomorrow" and "London"

​
Session Management

​
Creating Sessions

Sessions are created automatically when clients connect: 
# Server side - each connection gets a session
from kuralit.server.websocket_server import create_app

app = create_app(
    api_key_validator=lambda key: key == "demo-api-key",
    agent_session=agent,
)
# Each WebSocket connection = new session

​
Client Side

// Flutter client
final sessionId = Kuralit.generateSessionId();
await Kuralit.connect();
Kuralit.sendText(sessionId, 'Hello!');
// All messages with same sessionId share context

​
Best Practices

​
Context Length

  • Keep conversations focused - Long conversations may exceed context limits
  • Use tools effectively - Tools can help manage context by storing information
  • Clear session boundaries - Start new sessions for new topics

​
Context Quality

  • Clear user messages - Help the agent understand intent
  • Relevant tool results - Tools should return useful information
  • Consistent terminology - Use consistent terms throughout conversation

​
Next Steps