Pinecone Block

The Pinecone block lets you store, search, and retrieve vector embeddings using Pinecone's specialized vector database. Generate embeddings from text and perform semantic similarity searches with customizable filtering and reranking options. Reach for it when your workflow needs fast vector storage and semantic retrieval.

Overview

When to Use

• Generate dense or sparse embeddings from text using Pinecone's hosted embedding models.
• Upsert text records into a Pinecone index with automatic embedding via the Inference API.
• Run semantic similarity searches over an index by passing raw text (Pinecone embeds it on the fly).
• Query an index directly with a precomputed dense vector.
• Fetch specific vectors by their IDs from any namespace.
• Apply metadata filters and reranking to refine search results.

Configuration

Operation

Required dropdown selecting the action to perform. Every other field depends on which operation is chosen.

Model

Shown for generate. Dropdown selecting the Pinecone hosted embedding model.

Default: multilingual-e5-large.

Text Inputs

Shown for generate. Required. A JSON array of text objects to embed, e.g. [{"text": "Your text here"}].

Index Host

Shown for upsert_text, search_text, search_vector, and fetch. Required. The full Pinecone index host URL, e.g. https://index-name-abc123.svc.project-id.pinecone.io.

Namespace

Shown for upsert_text, search_text, search_vector, and fetch. Required. The namespace within the index to target (e.g. default).

Records

Shown for upsert_text. Required. NDJSON-formatted records to insert or update, each line containing _id, a text field (text or chunk_text), and optional metadata fields.

Search Query

Shown for search_text. Required. The natural-language text to search for. Pinecone embeds it automatically before querying.

Top K Results

Shown for search_text and search_vector. Optional. Number of nearest-neighbor results to return (default: 10).

Fields to Return

Shown for search_text. Optional. A JSON array of metadata field names to include in each result, e.g. ["category", "text"].

Filter

Shown for search_text. Optional. A JSON metadata filter to restrict search results, e.g. {"category": "product"}.

Rerank Options

Shown for search_text. Optional. A JSON object with reranking configuration, e.g. {"model": "bge-reranker-v2-m3", "rank_fields": ["text"], "top_n": 2}.

Query Vector

Shown for search_vector. Required. The dense query vector as a JSON array of floats, e.g. [0.1, 0.2, 0.3, ...].

Options

Shown for search_vector. A checkbox list of optional flags controlling what is returned in the response.

Vector IDs

Shown for fetch. Required. A JSON array of vector IDs to retrieve, e.g. ["vec1", "vec2"].

API Key

Required for all operations. Your Pinecone API key. Use {{PINECONE_API_KEY}} to pull it from an environment variable.

Inputs & Outputs

• Inputs:

  • operation (string) — Operation to perform
  • apiKey (string) — Pinecone API key
  • indexHost (string) — Index host URL
  • namespace (string) — Vector namespace
  • model (string) — Embedding model (for generate)
  • inputs (json) — Text inputs to embed (for generate)
  • parameters (json) — Model parameters (for generate)
  • records (json) — Records to upsert (for upsert_text)
  • searchQuery (string) — Search query text (for search_text)
  • topK (string) — Top K results (for search_text and search_vector)
  • fields (json) — Fields to return (for search_text)
  • filter (json) — Search filter (for search_text)
  • rerank (json) — Rerank options (for search_text)
  • ids (json) — Vector identifiers (for fetch)
  • vector (json) — Query vector (for search_vector)
  • includeValues (boolean) — Include vector values in response (for search_vector)
  • includeMetadata (boolean) — Include metadata in response (for search_vector)

• Outputs:

  • matches (json) — Search matches (from search_text, search_vector, and fetch)
  • upsertedCount (number) — Number of records successfully upserted (from upsert_text)
  • data (json) — Response data (from generate and fetch)
  • model (string) — Model used for generating embeddings (from generate)
  • vector_type (string) — Type of vector generated: dense or sparse (from generate)
  • usage (json) — Usage statistics (tokens, read units, rerank units)

Tools

• Pinecone Generate Embeddings (pinecone_generate_embeddings) — Calls the Pinecone Inference API (POST /embed) to generate dense or sparse embeddings from an array of text inputs using a hosted model.
• Pinecone Upsert Text (pinecone_upsert_text) — Inserts or updates text records in a Pinecone index via NDJSON upsert (POST /records/namespaces/{namespace}/upsert). Pinecone automatically embeds the text field.
• Pinecone Search Text (pinecone_search_text) — Searches a Pinecone index by natural-language text (POST /records/namespaces/{namespace}/search), with optional metadata filtering and reranking.
• Pinecone Search Vector (pinecone_search_vector) — Queries a Pinecone index with a raw dense vector (POST /query), returning nearest neighbors with optional values and metadata.
• Pinecone Fetch (pinecone_fetch) — Retrieves specific vectors by ID from a Pinecone index (GET /vectors/fetch), returning their values and metadata.

YAML Example

pinecone_1:
  type: pinecone
  name: "Pinecone"
  inputs:
    operation: "search_text"
    apiKey: "{{PINECONE_API_KEY}}"
    indexHost: "https://index-name-abc123.svc.project-id.pinecone.io"
    namespace: "default"
    searchQuery: "{{start.input}}"
    topK: "10"
    filter: '{"category": "product"}'
  connections:
    outgoing:
      - target: next-block-id