> ## Documentation Index
> Fetch the complete documentation index at: https://docs.findable.net/llms.txt
> Use this file to discover all available pages before exploring further.

# Vector Store Integration

> Query external vector stores as part of multi-source RAG or from inside Flow Designer workflows.

Findable provides two paths for querying external vector stores:

1. **Vector Retriever data source** (`type: 'vectorretriever'` in `dataSources[]`) — configure directly on a chat to query a vector store as part of multi-source RAG
2. **Flow Designer tool nodes** (`integrationType: 'vector_store'`) — use a vector store as a tool inside a Flow Designer workflow

Both approaches share the same `IDataConnection` registry for managing credentials and endpoints.

### Supported Providers

All vector stores are first-class `DataPlatformType` values in `IDataConnection` (see `VECTOR_STORE_TYPES` in `dataConnectionTypes.ts`).

**Tier 1 — Core:**

| Provider                                              | Platform Type   | Description                               |
| ----------------------------------------------------- | --------------- | ----------------------------------------- |
| [Pinecone](https://app.pinecone.io/)                  | `pinecone`      | Managed vector database                   |
| [pgvector](https://github.com/pgvector/pgvector)      | `pgvector`      | PostgreSQL vector extension               |
| [Weaviate](https://weaviate.io/)                      | `weaviate`      | Open-source vector database               |
| [Qdrant](https://qdrant.tech/)                        | `qdrant`        | High-performance vector similarity engine |
| [Chroma](https://www.trychroma.com/)                  | `chroma`        | Open-source embedding database            |
| [Elasticsearch](https://www.elastic.co/elasticsearch) | `elasticsearch` | Full-text + vector search                 |
| [OpenSearch](https://opensearch.org/)                 | `opensearch`    | AWS-managed search and analytics          |
| [Redis](https://redis.io/)                            | `redis`         | In-memory vector search                   |

**Tier 2 — Enterprise / Cloud:**

| Provider                                                                                  | Platform Type      | Description                  |
| ----------------------------------------------------------------------------------------- | ------------------ | ---------------------------- |
| [Milvus](https://milvus.io/)                                                              | `milvus`           | Open-source vector database  |
| [MongoDB Atlas](https://www.mongodb.com/products/platform/atlas-vector-search)            | `mongodb_atlas`    | Atlas Vector Search          |
| [Vertex AI Vector Search](https://cloud.google.com/vertex-ai/docs/vector-search/overview) | `vertex_ai_search` | Google Cloud managed vectors |
| [Amazon Kendra](https://aws.amazon.com/kendra/)                                           | `aws_kendra`       | Managed enterprise search    |
| [Oracle Vector Search](https://www.oracle.com/database/vector-database/)                  | `oracle_vector`    | Oracle Database 23ai vectors |

**Tier 3 — Specialized:**

| Provider                        | Platform Type | Description                    |
| ------------------------------- | ------------- | ------------------------------ |
| [Vespa](https://vespa.ai/)      | `vespa`       | Search + vector serving engine |
| [LanceDB](https://lancedb.com/) | `lancedb`     | Embedded vector database       |
| [Marqo](https://www.marqo.ai/)  | `marqo`       | End-to-end vector search       |

### Data Connection Fields (Vector Stores)

Vector store connections are managed via the unified `IDataConnection` interface in the `dataconnections` Cosmos container. When `type` is a vector store platform, these additional fields apply:

| Field              | Type       | Description                                                                         |
| ------------------ | ---------- | ----------------------------------------------------------------------------------- |
| `index`            | `string?`  | Index / collection / class name (Pinecone index, Weaviate class, Qdrant collection) |
| `namespace`        | `string?`  | Namespace / partition (Pinecone namespace, Redis prefix)                            |
| `verifySSL`        | `boolean?` | Verify SSL certificates (default: true)                                             |
| `maxRetries`       | `number?`  | Maximum retry attempts                                                              |
| `connectionString` | `string?`  | Full connection string (alternative to individual fields)                           |

Authentication types vary by provider — see `getAuthTypesForPlatform()` in `dataConnectionTypes.ts`.

### Vector Retriever Data Source

To query a vector store directly from a chat's RAG pipeline, add a `vectorretriever` entry to the chat's `dataSources[]`:

| Field                | Type                                              | Description                                              |
| -------------------- | ------------------------------------------------- | -------------------------------------------------------- |
| `type`               | `'vectorretriever'`                               | Identifies this as a vector retriever source             |
| `vectorConnectionId` | `string`                                          | FK to `IDataConnection` in the data connections registry |
| `vectorTopK`         | `number?`                                         | Number of results (default: 5)                           |
| `vectorSearchType`   | `'vector' \| 'semantic' \| 'hybrid' \| 'keyword'` | Search strategy                                          |
| `vectorMinScore`     | `number?`                                         | Minimum similarity score (0–1)                           |

The orchestrator queries the vector store in parallel with other data sources and merges results into the RAG context.

### Flow Designer Vector Store Config (`IVectorStoreConfig`)

For Flow Designer tool nodes with `integrationType: 'vector_store'`, the configuration has three parts:

**Connection** (`IVectorStoreConnectionConfig`):

* `endpoint`, `apiKey`, `index`, `environment`, `projectId`, `region`
* SSL/TLS, connection pool settings

**Search** (`IVectorStoreSearchConfig`):

* `topK`, `minScore`, `searchType` (vector / semantic / hybrid / keyword)
* `similarityMetric` (cosine / euclidean / dotProduct / l2)
* Metadata filters, namespace filters
* Reranking support (`enableReranking`, `rerankingModel`)
* Query expansion (`enableQueryExpansion`, `expansionPrompt`)
* Hybrid weights (`keywordWeight`, `vectorWeight`)

**Embedding** (optional):

* `modelEndpointId` — FK to AI model endpoint for query vectorization
* `embeddingDimensions`, `batchSize`

See [Vector Store Quick Reference](/docs/vector-store-quick-reference.md) and [Provider Template](/docs/vector-store-provider-template.md) for detailed setup.