Vector stores

The four vector store backends Aimogen Pro supports, how to configure each, and which to choose.

A vector store holds your embeddings and answers similarity queries. Aimogen Pro supports four.

Settings › Embeddings › Default Embeddings API To Use

Each option is only selectable once its credentials are configured — the settings screen disables the rest and says why.

Comparison#

OpenAI Vector StoresPineconeQdrantChroma
Added in2.8.3long-standinglong-standing2.8.3
HostingOpenAIManaged cloudManaged or self-hostedManaged or self-hosted
Setup effortLowestLowMediumMedium
Requires an OpenAI keyyesnonono
Self-hostablenonoyesyes
ChunkingHandled by OpenAIYou supply vectorsYou supply vectorsYou supply vectors

OpenAI Vector Stores#

The simplest option if you already use OpenAI. Files are uploaded and OpenAI handles chunking, embedding and retrieval.

SettingWhat it does
OpenAI Vector StoresSelect up to 2 stores to search
Create New StoreCreates one from the settings screen
OpenAI Vector Store Results LimitMaximum chunks returned per query. Default 3, range 1–50
OpenAI Vector Store Minimum Score (%)Minimum relevance score. 0 disables filtering. Recommended 20

The minimum score is worth setting. Without it, a query with no good match still returns the least-bad chunks, and the model treats them as relevant. A threshold makes "I do not have that information" possible.

If no store is configured, the plugin creates one named "Aiomatic Vector Store" on first use and records its ID.

Pinecone#

SettingWhat it does
Pinecone.io API KeyOn Settings › API Keys
Pinecone IndexThe index host, for example mytestingindex-28cc276.svc.us-east1-gcp.pinecone.io
Pinecone NamespaceOptional default namespace
Number Of Results To QueryChunks returned per query. Default 1

Number Of Results To Query defaults to 1, which is usually too few. Three to five gives the model enough context without flooding the prompt.

Qdrant#

SettingWhat it does
Qdrant API KeyOn Settings › API Keys. From the Data Access Control tab of your Qdrant dashboard
Qdrant Index URLFor example 29331ee6-....europe-west3-0.gcp.cloud.qdrant.io:6333
Qdrant Collection NameThe collection

Same requirement: dimension 1536, metric cosine.

Qdrant can be self-hosted, which matters if your embedded content must not leave your infrastructure.

Chroma#

SettingWhat it does
Chroma API KeyOn Settings › API Keys
Chroma Index URLFor example https://api.trychroma.com
Chroma TenantYour tenant
Chroma DatabaseThe database
Chroma Collection NameThe collection

Same dimension and metric requirements. Also self-hostable.

Choosing#

SituationChoose
Already using OpenAI, want the least setupOpenAI Vector Stores
Want a managed service with good toolingPinecone
Data must stay on your infrastructureQdrant or Chroma, self-hosted
Want to avoid an OpenAI dependencyPinecone, Qdrant or Chroma
EvaluatingOpenAI Vector Stores, then migrate if needed

Migrating between stores#

There is no migration tool. Changing store means re-indexing everything from source. That is one reason to keep your source content in a form you can re-index — the aiomatic_embeddings posts remain in WordPress, so re-indexing is feasible.

Cost#

OpenAI Vector Stores charge for storage per day plus retrieval.

Pinecone charges by index size and query volume, with a free tier.

Qdrant and Chroma charge for the managed service, or cost you a server if self-hosted.

For a typical WordPress site — a few thousand documents — all four are inexpensive.

Common problems#

Dimension mismatch errors. The index dimension does not match the embedding model. Recreate the index at the right dimension and re-index.

No results returned. The namespace does not match, the index is empty, or the score threshold is too high.

Connection errors. Wrong URL or key. Check the exact host format shown in the settings placeholders.

The store option is greyed out. Its credentials are not configured.

Still stuck? Open a support ticket and include the diagnostics from Aimogen Pro › System & Logs › System Info.