Overview
Pinecone lets a cloud flow store embeddings in your own Pinecone index and find the closest matches, e.g. to answer questions from your documents. The vectors stay in your account; Automize keeps nothing.
- Regions: Any Pinecone index, serverless or pod-based, in any cloud region.
- Plans: Cloud flows only. The bot has no Pinecone connector yet.
Setting up the connection
Auth type: API Key.
- In the Pinecone console, open API keys and create a key.
- Open the index and copy its Host, e.g. my-index-abc123.svc.aped-1234.pinecone.io. Each index has its own host, so make one connection per index.
- In Automize, open Settings → Connectors → Pinecone → Add connection, and paste the host and the key.
Rate limits
Set by Pinecone per index and plan. An upsert takes up to 1000 vectors.
Data model
Index (its own host) ─── Namespace ─── Vector
{"id": "doc-1", "values": [0.1, 0.2, …], "metadata": {"source": "invoice.pdf"}}Available operations
5 operations available. Click any row to jump to its detail.
| Operation | Category | What it does |
|---|---|---|
| Delete vectors | General | Delete vectors by id, or every vector in the namespace. |
| Get a vector | General | Read one vector, with its values and metadata, by its id. |
| Find similar vectors | General | Find the vectors closest to the one given. |
| Search by text | General | Search an index that turns text into vectors itself (one made with an embedding model in Pinecone), so no embedding step is needed first. |
| Upsert vectors | General | Add vectors to the index, or replace the ones with the same id. |
Operations
Delete vectors
Delete vectors by id, or every vector in the namespace.
| Parameter | Required | Type | Description |
|---|---|---|---|
Vector ids ids | No | list | The ids to delete, e.g. ["doc-1", "doc-2"]. |
Delete everything deleteAll | No | boolean | Delete every vector in the namespace. Leave Vector ids empty when this is on. |
Filter filter | No | object | Delete the vectors whose metadata matches (pod-based indexes only). |
Namespace namespace | No | text | The namespace inside the index. Leave empty for the default one. |
Get a vector
Read one vector, with its values and metadata, by its id.
| Parameter | Required | Type | Description |
|---|---|---|---|
Vector id ids | Yes | text | The id of the vector to read. |
Namespace namespace | No | text | The namespace inside the index. Leave empty for the default one. |
Tips
- One id per step.
Find similar vectors
Find the vectors closest to the one given.
| Parameter | Required | Type | Description |
|---|---|---|---|
Vector vector | Yes | list | The embedding to match, e.g. [0.1, 0.2, …]. It must have as many numbers as the index. |
How many topK | No | integer | How many matches to bring back. |
Filter filter | No | object | Only vectors whose metadata matches, e.g. {"source": {"$eq": "invoice.pdf"}}. |
Include metadata includeMetadata | No | boolean | Bring back each match's metadata. |
Namespace namespace | No | text | The namespace inside the index. Leave empty for the default one. |
Tips
- The matches are in matches, closest first, each with its score.
Search by text
Search an index that turns text into vectors itself (one made with an embedding model in Pinecone), so no embedding step is needed first.
| Parameter | Required | Type | Description |
|---|---|---|---|
Namespace namespace | Yes | text | The namespace to search. "__default__" is the default one. |
Text text | Yes | textarea | What to search for, in plain words. |
How many top_k | No | integer | How many matches to bring back. |
Filter filter | No | object | Only records whose fields match, e.g. {"category": {"$eq": "invoices"}}. |
Fields fields | No | list | The record fields to bring back, e.g. ["chunk_text", "source"]. Leave empty for all. |
Tips
- Only for an index made with an embedding model in Pinecone. Other indexes need Find similar vectors with an embedding from an AI step.
Upsert vectors
Add vectors to the index, or replace the ones with the same id.
| Parameter | Required | Type | Description |
|---|---|---|---|
Vectors vectors | Yes | list | A list of {"id": "doc-1", "values": [0.1, 0.2, …], "metadata": {"source": "invoice.pdf"}}. Up to 1000 per call. |
Namespace namespace | No | text | The namespace inside the index. Leave empty for the default one. |
Tips
- Every vector needs as many numbers as the index was made with.
FAQ
- Where do the vectors come from?
- From an embedding step, such as OpenAI's, earlier in the flow. An index with Pinecone's own embedding model can take text instead: use Search by text.