Cognee Community Vector Adapter - Moss

Use Moss as the vector database for Cognee.

Moss is a vector search service - indexes are built in the cloud and loaded locally for sub-10ms queries. No infrastructure to manage. Create an account, grab your credentials, and plug them in.

Prerequisites

1. Create a Moss account at portal.usemoss.dev
2. Create a project and copy your Project ID and Project Key
3. An OpenAI API key (Cognee uses it for LLM and embeddings by default)

Installation

pip install cognee-community-vector-adapter-moss

Or install from source:

cd packages/vector/moss
pip install -e .

Configuration

Set environment variables:

export MOSS_PROJECT_ID="your-project-id"
export MOSS_PROJECT_KEY="your-project-key"
export LLM_API_KEY="your-openai-api-key"

Or create a .env file (see .env.example).

Usage

import asyncio
import os
from cognee_community_vector_adapter_moss import register  # noqa: F401
from cognee import add, cognify, config, search


async def main():
    config.set_vector_db_config(
        {
            "vector_db_provider": "moss",
            "vector_db_key": os.getenv("MOSS_PROJECT_KEY"),
            "vector_db_name": os.getenv("MOSS_PROJECT_ID"),
            "vector_dataset_database_handler": "moss",
        }
    )

    await add("Natural language processing is a subfield of computer science.")
    await cognify()

    results = await search(query_text="Tell me about NLP")
    for r in results:
        print(r)


asyncio.run(main())

How It Works

The adapter multiplexes all Cognee collections into a single Moss index (cognee-index-{timestamp}). This keeps usage within Moss's free tier (3 indexes). Each document is tagged with a _collection metadata field, and all searches filter by it.

• Embeddings are computed by Cognee's embedding engine and passed to Moss via DocumentInfo(embedding=[...])
• The index is loaded locally via load_index(auto_refresh=True) for fast sub-10ms queries
• Async Moss jobs (create_index, add_docs) are polled to completion internally

Running Tests

MOSS_PROJECT_ID="..." MOSS_PROJECT_KEY="..." LLM_API_KEY="..." python tests/test_moss.py

Tests cover: vector search (text + vector), nodeset filtering, graph completion, chunks, summaries, and prune cleanup.

Resources

• Moss Documentation
• Moss Quickstart
• Cognee Documentation