Multi-turn conversations

Key Concepts: Maintaining Conversation Context

• API Limitation: The Anthropic API and Cloud do not store any messages (user inputs or model responses).
• Context Requirement: To achieve a multi-message conversation or maintain a flow, the user must manually manage the conversation history within their code.
• The Solution (Manual Context Management):
• Maintain a complete list of all messages exchanged.
• Provide this entire message list with every single follow-up request made to the API.
• Conversation Flow (The Process):
• Start with an empty message list.
• Add the initial user query to the list.
• Send the list to the API (using a chat function) to receive a response.
• Append the model's response (the assistant message) to the list.
• Add the next user query (the follow-up) to the list.
• Send the entire updated list back to the API to receive a contextual response.
• Helper Functions (Implementation): Effective conversation management relies on helper functions to structure the history:
• add user message: Appends a user's query (role: user) to the history list.
• add assistant message: Appends the model's response (role: assistant) to the history list.
• chat: The primary function that takes the full message history list and calls the API.

Takeaways

• The Anthropic API is stateless and does not store conversation history.
• To maintain context across multiple messages, the user must manually manage the conversation history within their code.
• Every follow-up request to the API must include the complete list of all previous user and assistant messages.
• Conversation flow requires appending both the user's query and the model's response to the history list before sending the entire list back to the API.