What is Pinecone Nexus?

Pinecone's knowledge engine for AI agents. It compiles documents, databases, apps and code into a structured knowledge layer that agents query in one call, with citations, confidence scores and access controls.

How is Nexus different from a vector database?

A vector database returns the closest chunks for each query. Nexus compiles and organizes knowledge ahead of time and enforces permissions, so agents get usable context rather than raw fragments.

Do we need Nexus, or is pgvector enough?

For one app over a moderate set of documents already near Postgres, pgvector is often enough. Nexus earns its place when many agents need governed access to many sources.

How do permissions work?

Access controls are applied in the knowledge layer based on the source permissions and the identity of the person or agent asking, so results differ by who is asking.

How do you measure whether it works?

We build a test set of real questions with known answers and sources, then measure correctness, citation accuracy and permission leaks before launch.

[ AI tools · Frameworks and platforms ]

Governed Knowledge for Agents, in One Call

Pinecone Nexus compiles your company’s documents, databases and apps into a knowledge layer agents can query with access controls and citations. We set it up, and build on Pinecone’s vector database where that fits better.

Plan Your Knowledge Layer Pinecone or pgvector?

SharePoint
Salesforce
Postgres
Jira
S3 bucket
GitHub

Compiled answer for the agent

Acme renews on 1 March. Two open P1 tickets mention the export bug; the fix ships in release 4.12.

Salesforce: Acme opportunity
Jira: SUP-2201, SUP-2207
GitHub: release 4.12 notes

Filtered to the asking user’s accessConfidence 0.92

Illustrative. One call returns grounded context instead of raw chunks.

Stop Making Every Agent Rediscover Your Data

Classic retrieval fetches raw chunks on every request and leaves the agent to sort them out. Nexus compiles knowledge ahead of time. Add agents and watch the difference.

Agents using your knowledge3

Retrieve every time

18 raw searches per round of questions (illustrative)

Search raw chunks on each request
Agent stitches fragments together
Permissions checked in your own code
Costs and latency grow with every agent

Compile once with Nexus

3 context calls to one compiled layer

Sources compiled into structured knowledge
One call returns grounded, cited context
Access controls enforced in the layer
Shared by every agent that needs it

[ Sources ]

Six Kinds of Data, One Knowledge Layer

Nexus runs in your own AWS, Google Cloud or Azure account and works with the models you choose.

01DocumentsPDFs, Office files and wikis from SharePoint, Google Drive and Confluence.
02DatabasesTables in Postgres, Snowflake and BigQuery, queried with their schema understood.
03Business appsRecords from CRM, ticketing and HR systems through connectors.
04Object storageFiles in S3, Cloud Storage or Azure Blob, kept in sync.
05CodeRepositories and their documentation for engineering agents.
06APIsLive data from your own services when compiled knowledge is not enough.

[ Governance ]

The Same Question, Three Different Correct Answers

Access follows each source’s permissions, so an agent acting for a person only sees what that person may see.

Sales rep
- Own accounts
- Pricing guide
- Other reps' deals
- HR files
Support agent
- Help center
- Ticket history
- Contracts
- Salary data
Finance analyst
- Invoices
- Revenue tables
- Medical leave records

[ Choosing ]

Nexus, Pinecone or pgvector?

We recommend the smallest thing that does the job.

01
pgvector in Postgres
Small to medium corpora, one app, data already in Postgres. Nothing new to run, back up or pay for.
Up to tens of millions of vectors
02
Pinecone vector database
Large or fast-growing corpora, many tenants, low-latency search managed for you.
Hybrid search, namespaces, serverless
03
Pinecone Nexus
Many agents and many sources that need shared, governed, compiled knowledge.
Context in one call, citations, access control

Where you sitpgvector in PostgresUp to tens of millions of vectorsOne app, small corpusMany agents, many sources

Hand pulling an index card from a wooden library card catalog

[ Knowledge audit ]

Map What Your Agents Should Know, and Who May See It

In a short call we list your sources, the permissions on each and the questions agents must answer, then propose the simplest setup that works.

Book the Audit

Start small with an assistant on pgvector or Pinecone AI Chatbot Starter · $1,500 to $2,500 , connect one agent to governed knowledge AI Agent Integration · $4,000 to $6,000 , or put Nexus in your cloud for every team Enterprise knowledge layer · from $8,000.

Plan Your Knowledge Layer

Packages from $1,500; custom projects from $8,000. We recommend the smallest thing that does the job.

Pinecone Questions

Pinecone's knowledge engine for AI agents. It compiles documents, databases, apps and code into a structured knowledge layer that agents query in one call, with citations, confidence scores and access controls.
A vector database returns the closest chunks for each query. Nexus compiles and organizes knowledge ahead of time and enforces permissions, so agents get usable context rather than raw fragments.
Pinecone states that Nexus can run inside your own AWS, Google Cloud or Azure account and works with the models you choose, including open-weight models.
For one app over a moderate set of documents already near Postgres, pgvector is often enough. Nexus earns its place when many agents need governed access to many sources.
Access controls are applied in the knowledge layer based on the source permissions and the identity of the person or agent asking, so results differ by who is asking.
We build a test set of real questions with known answers and sources, then measure correctness, citation accuracy and permission leaks before launch.