Backends

Medha supports nine vector backends. Choose based on your infrastructure, persistence requirements, and scale.

---

Comparison

Backend	Use Case	Persistence	Setup	Cloud-native	Extra
memory	Dev/testing	No	None	No	built-in
qdrant	General production	Yes	Docker/Cloud	Yes	[qdrant]
pgvector	Existing PostgreSQL	Yes	Postgres + extension	No	[pgvector]
elasticsearch	Search-heavy stacks	Yes	ES 8.x	Yes	[elasticsearch]
vectorchord	High-perf Postgres	Yes	Postgres + extension	No	[vectorchord]
chroma	Rapid prototyping	Optional	None/Docker	No	[chroma]
weaviate	Knowledge graph	Yes	Docker/Cloud	Yes	[weaviate]
redis	Low-latency cache	Optional	Redis Stack	Yes	[redis]
azure_search	Azure ecosystem	Yes	Managed	Yes	[azure-search]
lancedb	Embedded/serverless	Yes	None	Optional	[lancedb]

---

memory — In-Memory

Install

Built-in. No extra dependencies required.

pip install medha-archai

The default backend. Stores all vectors in a Python dict keyed by entry ID. Data is lost when the process exits. Ideal for unit tests and local development.

from medha import Settings

settings = Settings(backend_type="memory")

---

qdrant — Qdrant

Install

pip install "medha-archai[qdrant]"

Qdrant is the recommended production backend. It supports both an in-process mode (no server) and a full client-server deployment with gRPC/REST.

from medha import Settings

# Local Docker instance
settings = Settings(
    backend_type="qdrant",
    qdrant_host="localhost",
    qdrant_port=6333,
)

# Qdrant Cloud
settings = Settings(
    backend_type="qdrant",
    qdrant_url="https://your-cluster.qdrant.tech",
    qdrant_api_key="your-api-key",
)

Start a local Qdrant instance with Docker:

docker run -p 6333:6333 -p 6334:6334 qdrant/qdrant

---

pgvector — PostgreSQL + pgvector

Install

pip install "medha-archai[pgvector]"

Uses the pgvector extension for PostgreSQL. Best choice if you already run Postgres and want to consolidate infrastructure.

from medha import Settings

settings = Settings(
    backend_type="pgvector",
    pg_dsn="postgresql://user:password@localhost:5432/mydb",
)

Install the extension in your database:

CREATE EXTENSION IF NOT EXISTS vector;

---

elasticsearch — Elasticsearch

Install

pip install "medha-archai[elasticsearch]"

Uses Elasticsearch 8.x dense vector fields and kNN search. Suitable for stacks already using Elasticsearch for full-text search.

from medha import Settings

settings = Settings(
    backend_type="elasticsearch",
    es_url="http://localhost:9200",
    # es_api_key="..." for Elastic Cloud
)

---

vectorchord — VectorChord

Install

pip install "medha-archai[vectorchord]"

VectorChord is a high-performance vector extension for PostgreSQL using IVF-PQ indexing. Recommended over pgvector for large-scale deployments (>1M entries).

from medha import Settings

settings = Settings(
    backend_type="vectorchord",
    pg_dsn="postgresql://user:password@localhost:5432/mydb",
)

---

chroma — Chroma

Install

pip install "medha-archai[chroma]"

Chroma can run in-process (ephemeral) or as a server (persistent). Excellent for rapid prototyping and notebook environments.

from medha import Settings

# In-process (ephemeral)
settings = Settings(backend_type="chroma")

# HTTP client (persistent)
settings = Settings(
    backend_type="chroma",
    chroma_host="localhost",
    chroma_port=8000,
)

---

weaviate — Weaviate

Install

pip install "medha-archai[weaviate]"

Weaviate is a knowledge-graph-oriented vector database with built-in schema management. Good for teams already using Weaviate or who need rich object properties alongside vectors.

from medha import Settings

settings = Settings(
    backend_type="weaviate",
    weaviate_url="http://localhost:8080",
    # weaviate_api_key="..." for Weaviate Cloud
)

---

redis — Redis Stack

Install

pip install "medha-archai[redis]"

Uses Redis Stack's RediSearch module for vector similarity search. Delivers the lowest latency of any networked backend (< 1 ms P99 for small indexes). Persistence depends on your Redis configuration (RDB/AOF).

from medha import Settings

settings = Settings(
    backend_type="redis",
    redis_url="redis://localhost:6379",
)

Note

The redis backend uses Redis Stack, not vanilla Redis. Ensure RediSearch is available in your deployment.

---

azure_search — Azure AI Search

Install

pip install "medha-archai[azure-search]"

Fully managed vector search service on Azure. Recommended for teams running workloads in Azure that want zero infrastructure management.

from medha import Settings

settings = Settings(
    backend_type="azure_search",
    azure_search_endpoint="https://your-service.search.windows.net",
    azure_search_api_key="your-admin-key",
)

---

lancedb — LanceDB

Install

pip install "medha-archai[lancedb]"

LanceDB is an embedded vector database that writes directly to disk (local or cloud object storage). No server required, making it ideal for serverless deployments or single-machine applications that need persistence.

from medha import Settings

# Local filesystem
settings = Settings(
    backend_type="lancedb",
    lancedb_uri="./my_cache.lancedb",
)

# S3 / GCS / Azure Blob
settings = Settings(
    backend_type="lancedb",
    lancedb_uri="s3://my-bucket/cache",
)