Chat API

​

Overview

​

Send Chat Message

​

Request

​

Response

​

Example Request

curl -X POST https://api.decipherit.com/chat/message \
  -H "Content-Type: application/json" \
  -d '{
    "notebook_id": "nb_123456",
    "messages": [
      {
        "role": "user",
        "content": "What are the key findings about quantum computing?"
      }
    ]
  }'

​

Example Response

{
  "status": "success",
  "response": "Based on the research in your notebook, the key findings about quantum computing include: 1) Quantum computers leverage quantum superposition and entanglement to process information exponentially faster than classical computers for specific problems. 2) Current applications focus on cryptography, drug discovery, and optimization problems. 3) Major challenges include maintaining qubit coherence and error correction."
}

​

Multi-turn Conversation

curl -X POST https://api.decipherit.com/chat/message \
  -H "Content-Type: application/json" \
  -d '{
    "notebook_id": "nb_123456",
    "messages": [
      {
        "role": "user",
        "content": "What are the key findings?"
      },
      {
        "role": "assistant",
        "content": "The key findings include..."
      },
      {
        "role": "user",
        "content": "Can you elaborate on the first point?"
      }
    ]
  }'

​

Example with Metadata

{
  "notebook_id": "nb_123456",
  "messages": [
    {
      "role": "user",
      "content": "Summarize the main points"
    }
  ],
  "metadata": {
    "user_id": "user_789",
    "session_id": "session_xyz"
  }
}

​

Error Responses

{
  "detail": "Failed to process chat message"
}

​

How It Works

1. Vector Search: Your message is used to search the notebook’s research content using semantic similarity
2. Context Retrieval: Relevant research sections are retrieved from the vector database
3. AI Generation: The chat agent generates a response using the retrieved context and conversation history
4. Response: The generated answer is returned, grounded in your notebook’s research

​

Best Practices

• Be Specific: Ask specific questions about the research topic for more focused answers
• Maintain Context: Include previous messages for coherent multi-turn conversations
• Notebook Scope: Questions are answered based only on the specified notebook’s research content
• Error Handling: Implement retry logic for 500 errors as they may be transient