Memory Vault (ChromaDB)

The Memory Vault gives your agent persistent, searchable long-term memory using vector embeddings.

Overview

OpenZess uses ChromaDB — an open-source vector database — to store important context that persists across sessions.

Unlike regular chat history (which is stored in PostgreSQL), the Memory Vault uses semantic similarity search. This means the agent can find relevant memories based on meaning, not just keywords.

How It Works

import chromadb

chroma_client = chromadb.PersistentClient(path="./chroma_db")
memory_collection = chroma_client.get_or_create_collection(
    name="openzess_memory"
)

Storing Memories

The agent can store text snippets with metadata:

memory_collection.add(
    documents=["User prefers Python over JavaScript"],
    ids=["mem-001"],
    metadatas=[{"source": "conversation", "topic": "preferences"}]
)

Retrieving Memories

Retrieve memories based on semantic similarity:

results = memory_collection.query(
    query_texts=["What language does the user like?"],
    n_results=3
)

Managing Memories

The frontend provides a dedicated Memory Vault page where you can:

• View all memories with their stored text and metadata
• Delete individual memories that are no longer relevant
• Clear all memories to start fresh

API Endpoints

Method	Endpoint	Description
GET	/api/memory	List all stored memories
DELETE	/api/memory/{id}	Delete a specific memory
DELETE	/api/memory	Clear all memories