Pinecone

Pinecone é um banco de dados vetorial projetado para construir aplicações de busca vetorial de alto desempenho. Ele permite armazenamento, gerenciamento e busca por similaridade eficientes de embeddings vetoriais de alta dimensão, tornando-o ideal para aplicações de IA que exigem capacidades de busca semântica.

Com o Pinecone, você pode:

  • Armazenar embeddings vetoriais: Gerencie vetores de alta dimensão em escala de forma eficiente
  • Realizar busca por similaridade: Encontre os vetores mais semelhantes a um vetor de consulta em milissegundos
  • Construir busca semântica: Crie experiências de busca baseadas em significado em vez de palavras-chave
  • Implementar sistemas de recomendação: Gere recomendações personalizadas com base na similaridade de conteúdo
  • Implantar modelos de machine learning: Operacionalize modelos de ML que dependem de similaridade vetorial
  • Escalar de forma contínua: Lide com bilhões de vetores com desempenho consistente
  • Manter índices em tempo real: Atualize seu banco de dados vetorial em tempo real conforme novos dados chegam

Na Zoen, a integração com o Pinecone permite que seus agentes aproveitem capacidades de busca vetorial de forma programática como parte dos seus fluxos de trabalho. Isso habilita cenários sofisticados de automação que combinam processamento de linguagem natural com busca e recuperação semântica. Seus agentes podem gerar embeddings a partir de texto, armazenar esses vetores em índices do Pinecone e realizar buscas por similaridade para encontrar as informações mais relevantes. Esta integração conecta seus fluxos de IA à infraestrutura de busca vetorial, permitindo recuperação de informação mais inteligente com base em significado semântico em vez de correspondência exata de palavras-chave. Ao conectar a Zoen ao Pinecone, você pode criar agentes que entendem contexto, recuperam informações relevantes de grandes conjuntos de dados e entregam respostas mais precisas e personalizadas aos usuários — tudo sem exigir gerenciamento complexo de infraestrutura ou conhecimento especializado em bancos de dados vetoriais.

Usage Instructions

Integrate Pinecone into the workflow. Generate embeddings, upsert and update text records, delete vectors, search with text or vectors, fetch and list vectors, inspect index statistics, and manage indexes.

Actions

pinecone_generate_embeddings

Generate embeddings from text using Pinecone's hosted models

Input

ParameterTypeRequiredDescription
modelstringYesModel to use for generating embeddings
inputsarrayYesArray of text inputs to generate embeddings for
apiKeystringYesPinecone API key

Output

ParameterTypeDescription
dataarrayGenerated embeddings data with values and vector type
modelstringModel used for generating embeddings
vector_typestringType of vector generated (dense/sparse)
usageobjectUsage statistics for embeddings generation

pinecone_upsert_text

Insert or update text records in a Pinecone index

Input

ParameterTypeRequiredDescription
indexHoststringYesFull Pinecone index host URL (e.g., "https://my-index-abc123.svc.pinecone.io"\)
namespacestringYesNamespace to upsert records into (e.g., "documents", "embeddings")
recordsarrayYesRecord or array of records to upsert, each containing _id, text, and optional metadata
apiKeystringYesPinecone API key

Output

ParameterTypeDescription
statusTextstringStatus of the upsert operation

pinecone_update_vector

Update the values, sparse values, or metadata of a vector in a Pinecone namespace

Input

ParameterTypeRequiredDescription
indexHoststringYesFull Pinecone index host URL (e.g., "https://my-index-abc123.svc.pinecone.io"\)
idstringYesUnique ID of the vector to update
namespacestringNoNamespace containing the vector (e.g., "documents", "embeddings")
valuesarrayNoNew dense vector values to overwrite the existing values
sparseValuesobjectNoNew sparse vector values with indices and values arrays
setMetadataobjectNoMetadata key-value pairs to add or overwrite on the vector
apiKeystringYesPinecone API key

Output

ParameterTypeDescription
statusTextstringStatus of the update operation

pinecone_delete_vectors

Delete vectors from a Pinecone namespace by IDs, by metadata filter, or delete all

Input

ParameterTypeRequiredDescription
indexHoststringYesFull Pinecone index host URL (e.g., "https://my-index-abc123.svc.pinecone.io"\)
namespacestringNoNamespace to delete vectors from (e.g., "documents", "embeddings")
idsarrayNoVector IDs to delete (1-1000 items). Mutually exclusive with deleteAll and filter
deleteAllbooleanNoDelete all vectors in the namespace. Mutually exclusive with ids and filter
filterobjectNoMetadata filter selecting vectors to delete (e.g., {"category": {"$eq": "product"}}). Mutually exclusive with ids and deleteAll
apiKeystringYesPinecone API key

Output

ParameterTypeDescription
statusTextstringStatus of the delete operation

pinecone_search_text

Search for similar text in a Pinecone index

Input

ParameterTypeRequiredDescription
indexHoststringYesFull Pinecone index host URL (e.g., "https://my-index-abc123.svc.pinecone.io"\)
namespacestringNoNamespace to search in (e.g., "documents", "embeddings")
searchQuerystringYesText to search for
topKstringNoNumber of results to return (e.g., "10", "25")
fieldsarrayNoFields to return in the results
filterobjectNoFilter to apply to the search (e.g., {"category": "tech", "year": {"$gte": 2020}})
rerankobjectNoReranking parameters
apiKeystringYesPinecone API key

Output

ParameterTypeDescription
matchesarraySearch results with ID, score, and metadata
idstringVector ID
scorenumberSimilarity score
metadataobjectAssociated metadata
usageobjectUsage statistics including tokens, read units, and rerank units
total_tokensnumberTotal tokens used for embedding
read_unitsnumberRead units consumed
rerank_unitsnumberRerank units used

pinecone_search_vector

Search for similar vectors in a Pinecone index

Input

ParameterTypeRequiredDescription
indexHoststringYesFull Pinecone index host URL (e.g., "https://my-index-abc123.svc.pinecone.io"\)
namespacestringNoNamespace to search in (e.g., "documents", "embeddings")
vectorarrayYesVector to search for
topKnumberNoNumber of results to return (e.g., 10, 25)
filterobjectNoFilter to apply to the search (e.g., {"category": "tech", "year": {"$gte": 2020}})
includeValuesbooleanNoInclude vector values in response
includeMetadatabooleanNoInclude metadata in response (true/false)
apiKeystringYesPinecone API key

Output

ParameterTypeDescription
matchesarrayVector search results with ID, score, values, and metadata
namespacestringNamespace where the search was performed

pinecone_fetch

Fetch vectors by ID from a Pinecone index

Input

ParameterTypeRequiredDescription
indexHoststringYesFull Pinecone index host URL (e.g., "https://my-index-abc123.svc.pinecone.io"\)
idsarrayYesArray of vector IDs to fetch (e.g., ["vec-001", "vec-002"])
namespacestringNoNamespace to fetch vectors from (e.g., "documents", "embeddings")
apiKeystringYesPinecone API key

Output

ParameterTypeDescription
matchesarrayFetched vectors with ID, values, metadata, and score
idstringVector ID
valuesarrayVector values
metadataobjectAssociated metadata
scorenumberMatch score (1.0 for exact matches)
dataarrayVector data with values and vector type
valuesarrayVector values
vector_typestringVector type (dense/sparse)
usageobjectUsage statistics including total read units
total_tokensnumberRead units consumed

pinecone_list_vector_ids

List vector IDs in a Pinecone namespace by prefix (serverless indexes only)

Input

ParameterTypeRequiredDescription
indexHoststringYesFull Pinecone index host URL (e.g., "https://my-index-abc123.svc.pinecone.io"\)
namespacestringYesNamespace to list vector IDs from (e.g., "documents", "embeddings")
prefixstringNoFilter vector IDs by a common prefix (e.g., "doc1#")
limitnumberNoMaximum number of IDs to return per page (default 100)
paginationTokenstringNoPagination token from a previous response to fetch the next page
apiKeystringYesPinecone API key

Output

ParameterTypeDescription
vectorIdsarrayVector IDs in the namespace
paginationobjectPagination info with a next token when more results exist
nextstringToken to fetch the next page
namespacestringNamespace the IDs were listed from
usageobjectUsage statistics including read units
total_tokensnumberRead units consumed

pinecone_describe_index_stats

Get statistics about a Pinecone index, including per-namespace vector counts

Input

ParameterTypeRequiredDescription
indexHoststringYesFull Pinecone index host URL (e.g., "https://my-index-abc123.svc.pinecone.io"\)
filterobjectNoMetadata filter to limit which vectors are counted (pod-based indexes only, e.g., {"category": {"$eq": "product"}})
apiKeystringYesPinecone API key

Output

ParameterTypeDescription
namespacesjsonMap of namespace name to its summary including vectorCount
dimensionnumberDimensionality of the indexed vectors
indexFullnessnumberFullness of the index (pod-based indexes only)
totalVectorCountnumberTotal number of vectors across all namespaces

pinecone_list_indexes

List all Pinecone indexes in the project

Input

ParameterTypeRequiredDescription
apiKeystringYesPinecone API key

Output

ParameterTypeDescription
indexesarrayList of indexes with name, dimension, metric, host, spec, and status
namestringIndex name
dimensionnumberVector dimensionality
metricstringDistance metric (cosine, euclidean, dotproduct)
hoststringIndex host URL for data-plane operations
vectorTypestringVector type (dense or sparse)
deletionProtectionstringDeletion protection (enabled or disabled)
tagsobjectCustom user tags on the index
specobjectIndex spec (serverless or pod configuration)
statusobjectIndex status with ready and state

pinecone_describe_index

Get the configuration and status of a Pinecone index by name

Input

ParameterTypeRequiredDescription
indexNamestringYesName of the index to describe
apiKeystringYesPinecone API key

Output

ParameterTypeDescription
indexobjectIndex configuration and status
namestringIndex name
dimensionnumberVector dimensionality
metricstringDistance metric (cosine, euclidean, dotproduct)
hoststringIndex host URL for data-plane operations
vectorTypestringVector type (dense or sparse)
deletionProtectionstringDeletion protection (enabled or disabled)
tagsobjectCustom user tags on the index
specobjectIndex spec (serverless or pod configuration)
statusobjectIndex status with ready and state

On this page