Pinecone

Managed vector database for agent memory. Manage serverless indexes, upsert and query vectors, work with integrated-embedding text records, and call hosted models for embeddings and reranking.

19 actions available

A new document arrives and needs to enter the knowledge base

Agent reads the document text and splits it into chunks

Agent writes the chunks with Upsert Records into a namespace

Agent confirms freshness with Describe Index Stats

Agent runs Query Vectors to confirm the new content is retrievable

Agent posts the index and namespace update to the owning team

Any index or namespace deletion waits for an owner to approve

What This Integration Enables

The Pinecone connector lets agents own the full lifecycle of a vector store. Agents provision serverless indexes, write vectors or integrated-embedding text records into namespaces, and query for nearest neighbors with metadata filtering. Create Embeddings and Rerank Documents run against Pinecone-hosted models, so an agent can embed and rerank without a separate provider when that is simpler. Describe Index and Describe Index Stats let the agent check readiness and freshness before it acts on a result. Everything a RAG pipeline needs is a step, not a build.

Without FlowRunner

Knowledge scattered Reference material sits in folders and inboxes no agent can search

Retrieval hand-built Someone stands up embeddings and search infrastructure per project

Stale answers Agents answer from a snapshot that nobody keeps current

With FlowRunner

Searchable memory Documents become vectors an agent can query in one step

Retrieval as a capability Upsert and query are workflow actions, not a project

Memory stays current New content is embedded and indexed as it arrives

Use Case Scenarios

Knowledge Base That Stays In Sync

A support team keeps its answers in a set of documents that change every week. When a document changes, the agent chunks the text, calls Upsert Records into the collection namespace, then calls Describe Index Stats to confirm the write landed. Because the index uses integrated embedding, the agent does not run a separate embedding step. The knowledge base tracks the source material without anyone rebuilding it.

Grounded Answer With Reranking

A user asks a question. The agent embeds the question, calls Query Vectors for the top matches, then calls Rerank Documents to reorder them by relevance to the exact wording. It hands the top passages to an LLM as grounding context, records the retrieval quality in a tracking record, and returns an answer that cites its sources. Retrieval and reranking both run inside one flow.

Cleaning Up A Retired Project

A project ends and its namespace should be removed. The agent gathers the namespace record count with Describe Index Stats and the last query activity, assembles a removal summary, and routes it to the data owner. Only after approval does it call Delete Namespace. The cleanup is deliberate, not a reflex.

Human-in-Loop Highlight

Reading a vector store is safe. Destroying part of one is not. [Human-in-the-loop](/concepts/human-in-the-loop/) is an execution pattern where AI agents pause autonomously, assemble the relevant context and the decision choices available, route to a human via their preferred channel, and resume the moment the human responds. The Pinecone connector places that pause on the one action a query cannot undo: deleting an index or a namespace. When a flow reaches a Delete Index or Delete Namespace step, the agent does not run it silently. It gathers what the index holds, the vector count from Describe Index Stats, and the recent query activity, then asks the owner through their channel: "The retired-project namespace holds 84,000 vectors and was last queried nine days ago. Deleting it is permanent. Delete it?" The agent handles provisioning, embedding, and search on its own. A person owns every action that erases memory the team cannot get back.

Agent processes routinely

Detects exception requiring judgment

Clear match Continues automatically

Ambiguous Routes to human via preferred channel

Human decides

Agent resumes with decision

Agent Capabilities

19 actions

Indexes

Configure Index Update the configuration of an existing index, including toggling deletion protection and replacing custom tags.
Create Index Create a new serverless index with an explicit vector dimension and similarity metric in the specified cloud and region. Creation is asynchronous.
Create Index for Model Create a serverless index with integrated embedding, so records upserted as raw text are embedded automatically by a hosted model. Required for Upsert Records and Search Records.
Delete Index Permanently delete an index and all of its data. This cannot be undone and fails if deletion protection is enabled.
Describe Index Retrieve the full configuration and current status of an index, including its data-plane host. Check status.ready before using a new index.
List Indexes Retrieve all indexes in the project, including each index name, dimension, metric, host, status, and spec.

Vectors

Delete Vectors Delete vectors from a namespace by explicit IDs, by metadata filter, or all vectors in the namespace.
Describe Index Stats Return statistics about an index, including total vector count, index fullness, and a per-namespace breakdown.
Fetch Vectors Look up vectors by ID from a namespace and return their values and metadata.
List Vector IDs List the IDs of vectors in a namespace of a serverless index, optionally filtered by an ID prefix. Paginated.
Query Vectors Search a namespace for the vectors most similar to a query vector or an existing vector by ID, with metadata filtering. The core retrieval operation.
Update Vector Update an existing vector by ID, replacing its values and/or merging new metadata fields.
Upsert Vectors Write vectors into a namespace. If a vector ID already exists, its values and metadata are overwritten.

Records

Search Records Perform a semantic text search over an integrated-embedding index. The query text is embedded automatically, with optional hosted reranking.
Upsert Records Write text records into an integrated-embedding index; each record text field is embedded automatically by the hosted model. No separate embedding step needed.

Namespaces

Delete Namespace Permanently delete a namespace from a serverless index, including all vectors it contains. This cannot be undone.
List Namespaces List the namespaces in a serverless index along with each namespace record count. Paginated.

Inference

Create Embeddings Generate vector embeddings for input texts using a Pinecone-hosted embedding model. Set input type to Passage for documents or Query for search queries.
Rerank Documents Rerank a list of documents by relevance to a query using a Pinecone-hosted reranking model. Returns documents ordered by relevance score.

Start building with Pinecone

$100 in credits. No card required. Connect in minutes.

Start Building Free Talk to us