Skip to main content

​
POST /memory/add

Add new memories from a conversation. Mem0 automatically extracts and stores relevant facts from the messages. The API can automatically classify the memory type using LLM-based classification.

​
Request Body

​
messages
array
required
Array of message objects representing the conversation.
​
user_id
string
required
Unique identifier for the user. All memories are associated with this user ID.
​
metadata
object
Optional metadata to attach to the memory. You can include custom fields like tags, timestamps, or manually specify memory_type.
​
auto_classify
boolean
default:"true"
Enable automatic memory type classification. When enabled, the API uses GPT-4.1-nano to classify the memory into one of five types: LONG_TERM, SHORT_TERM, EPISODIC, SEMANTIC, or PROCEDURAL.

​
Response

​
success
boolean
Indicates whether the operation was successful.
​
result
object
The result from mem0 containing memory IDs and extracted information.
​
classified_type
string
The automatically classified memory type (LONG_TERM, SHORT_TERM, EPISODIC, SEMANTIC, or PROCEDURAL). Only present if auto_classify is enabled.

​
Example Request

curl -X POST http://localhost:8000/memory/add \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      {
        "role": "user",
        "content": "I prefer dark mode in all my applications"
      }
    ],
    "user_id": "user_123",
    "auto_classify": true
  }'

​
Example Response

{
  "success": true,
  "result": {
    "results": [
      {
        "id": "mem_abc123",
        "memory": "User prefers dark mode in applications",
        "event": "ADD"
      }
    ]
  },
  "classified_type": "LONG_TERM"
}

​
Manual Memory Type Override

You can manually specify the memory type by including it in the metadata: 
curl -X POST http://localhost:8000/memory/add \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      {
        "role": "user",
        "content": "Yesterday I attended the React conference"
      }
    ],
    "user_id": "user_123",
    "metadata": {
      "memory_type": "EPISODIC",
      "tags": ["conference", "react"]
    }
  }'

​
Error Responses

​
detail
string
Error message describing what went wrong.
500 Internal Server Error 
{
  "detail": "Failed to connect to vector store"
}

​
Implementation Details

  • Source: backend/main.py:189
  • Classification Model: GPT-4.1-nano-2025-04-14
  • Default Classification: LONG_TERM (used as fallback on classification errors)
  • Memory Extraction: Powered by mem0, which intelligently extracts facts from conversations

Build docs developers (and LLMs) love

Get started for freeTalk to us