Skip to main content

Memory

The platform includes semantic memory powered by Mem0, allowing conversations to remember information across sessions. Memories are per-user and can be searched, managed, and automatically injected into conversation context.

How Memory Works​

Memories are facts, preferences, and context extracted from conversations and stored in a vector database. When a new message is sent, relevant memories can be retrieved and included in the LLM's context, giving the AI long-term recall.

Searching Memories​

Search across a user's memories using natural language. Results are ranked by semantic similarity.

curl -X POST https://api.yocaso.dev/api/v1/llm/gateway/search-memories \
  -H "X-API-Key: sk_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "dietary preferences"
  }'

Response:

{
  "memories": [
    {
      "id": "mem_abc123",
      "memory": "User is vegetarian and avoids gluten",
      "score": 0.92,
      "created_at": "2025-02-15T10:00:00Z",
      "updated_at": "2025-02-15T10:00:00Z"
    },
    {
      "id": "mem_def456",
      "memory": "User prefers meals under 500 calories",
      "score": 0.85,
      "created_at": "2025-02-20T14:30:00Z",
      "updated_at": "2025-02-20T14:30:00Z"
    }
  ]
}

Listing Memories​

List all memories for the authenticated user (paginated):

curl -X POST https://api.yocaso.dev/api/v1/llm/gateway/list-memories \
  -H "X-API-Key: sk_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{}'

Getting a Specific Memory​

curl -X POST https://api.yocaso.dev/api/v1/llm/gateway/get-memory \
  -H "X-API-Key: sk_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{
    "memory_id": "mem_abc123"
  }'

Updating a Memory​

Correct or refine an existing memory:

curl -X POST https://api.yocaso.dev/api/v1/llm/gateway/update-memory \
  -H "X-API-Key: sk_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{
    "memory_id": "mem_abc123",
    "memory": "User is vegetarian, avoids gluten, and prefers organic produce"
  }'

Deleting Memories​

Delete a Single Memory​

curl -X POST https://api.yocaso.dev/api/v1/llm/gateway/delete-memory \
  -H "X-API-Key: sk_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{
    "memory_id": "mem_abc123"
  }'

Delete All Memories​

danger

This permanently deletes all memories for the user. This action cannot be undone.

curl -X POST https://api.yocaso.dev/api/v1/llm/gateway/delete-all-memories \
  -H "X-API-Key: sk_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{}'

Memory Timestamp Injection​

When memories are injected into conversation context, each memory can include a timestamp annotation showing when it was remembered (and optionally when it was last updated). This helps the LLM understand the recency and relevance of each memory.

Example context block:

[Relevant memories about this user:]
- User is vegetarian and avoids gluten (remembered: Feb 15, 2025 at 10:00 AM UTC)
- User prefers meals under 500 calories (remembered: Feb 20, 2025 at 2:30 PM UTC, updated: Mar 1, 2025 at 9:00 AM UTC)

Timestamp injection is enabled by default. To disable it, set inject_memory_timestamps to false in the Mem0 configuration:

{
  "mem0_config": {
    "enabled": true,
    "inject_memory_timestamps": false
  }
}
FieldTypeDefaultDescription
inject_memory_timestampsbooltrueInclude (remembered: <date>) annotations on injected memories. Adds (updated: <date>) if the memory was modified after creation.

Memory and Conversations​

Memories integrate with conversations in two ways:

  1. Automatic retrieval — When a message is sent, the platform can search for relevant memories and include them in the LLM's context, giving it awareness of past interactions.

  2. MCP tool access — Memory operations are exposed as MCP tools, allowing the LLM to search, create, and update memories during a conversation.

User Impersonation​

When using API keys with the users:impersonate scope, memory operations act on the impersonated user's memories:

curl -X POST https://api.yocaso.dev/api/v1/llm/gateway/search-memories \
  -H "X-API-Key: sk_your_key_here" \
  -H "X-On-Behalf-Of: user_123" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "exercise habits"
  }'