Vector Store

Cencori manages a high-performance vector database under the hood, so you don't have to manage Pinecone or Weaviate instances.

Embeddings

We automatically generate embeddings for all stored content using OpenAI's text-embedding-3-small model.

• Dimensions: 1536
• Max Tokens: 8191 per chunk
• Language: Multilingual support

[!NOTE] We automatically chunk long text fields. You can control chunk size in the store options (default: 1000 tokens).

Time To Live (TTL)

You can set memories to auto-expire after a certain duration. This is useful for:

• Session history (expire after 24h)
• Cached context (expire after 7 days)
• Temporary user data

await cencori.memory.store({
  namespace: 'session-123',
  content: 'User is interested in pricing',
  ttl: 86400 // Expire in 24 hours (seconds)
});

Indexing Latency

Cencori indexes new memories in <500ms. This "near real-time" availability means you can store a memory and search for it almost immediately in the next turn of conversation.

Import / Export

Importing Data

To migrate data from another vector store, format your data as a JSONL file and use our CLI import tool:

cencori memory import ./data.jsonl --namespace=docs

JSONL Format:

{"content": "...", "metadata": {"key": "value"}}
{"content": "...", "metadata": {"key": "value"}}

Exporting Data

You can export a snapshot of any namespace for backup or analysis.

cencori memory export --namespace=docs > backup.jsonl