Skip to main content
Pinecone Nexus runs in your own cloud account. This quickstart deploys Nexus with Bring Your Own Cloud (BYOC), turns a set of your documents into a context, and queries it for grounded, cited answers. See the overview for what you can do with Nexus. Nexus organizes knowledge in a hierarchy:
  • A workspace holds one or more contexts.
  • A context is built from the documents you ingest into it.
  • A query can read one or more contexts, so several contexts together can give an agent its knowledge.
See key concepts for the full model.
The examples in this quickstart use a company knowledge base of internal policies for illustration. You bring your own documents, and any set works.

​
Prerequisites

  • A Pinecone Enterprise plan (required for BYOC).
  • A dedicated cloud account (AWS, GCP, or Azure) with admin access, plus the install tooling. See the deploy prerequisites for the full list.
  • A Pinecone API key for the target project.
  • Documents to build your context from, such as files to upload or a Hugging Face, GitHub, Box, or Google Drive source.

​
1. Deploy Nexus

Nexus runs in your own cloud, so deploying it comes first.
1

Install Nexus

Follow Deploy Nexus BYOC to install Nexus in your own cloud account with the Pulumi installer.
2

Get your console URL and API host

When the install finishes, it prints your workspace console URL and API host. Open the console to work in the UI, or use the host as your data-plane base URL for the API. See Authentication to set NEXUS_BASE_URL and get a token.

​
2. Add a context

Build a context from your own documents. A company knowledge base might include a refund policy, an expense policy, and a vendor-approval matrix. Nexus ingests dense documents like PDFs directly, alongside Markdown and plain text. For example, two PDFs and a Markdown file could read:
  • policies/refunds.md: Refund requests must be submitted within 30 days of purchase. After that window, customers receive store credit, not a direct refund.
  • policies/expenses.pdf: Standard expenses over $1,000 require manager approval before reimbursement.
  • policies/vendors.pdf: Vendor invoices over $10,000 require finance approval. Invoices over $25,000 also require VP sign-off.
The queries later in this quickstart draw on these files. You can build the context through the API or the console.
1

Set your base URL and token

Point at your workspace host and exchange your Pinecone API key for a session token. See Authentication for details.
2

Create the context

Create the context with a slug and a name:
curl
A new context curates under the default manifest. To define your own artifact and edge types, see Design your own manifest.
3

Add sources

Upload each of your files, one request per file (archives are expanded automatically), or import from a connector with POST /contexts/{slug}/import:
curl
4

Curate

Curate the context to build its knowledge from the sources:
curl
Curation chunks your sources, distills them into typed artifacts, and indexes everything. It runs as a background task, so query the context once it finishes. See How curation works.

​
3. Query your context

Once curation finishes, ask your context a question and get back a grounded answer with citations.
Send a KnowQL query over HTTP, passing the context slug as scope:
curl
Output
The full response also carries citations and usage. See the Nexus API for the full surface, or connect an MCP server for Claude Desktop and other MCP clients.
To query several contexts at once, pass multiple slugs in the query’s scope array. In the console, start a session from Sessions with + New session and select each context to query across.

​
A grounded, cited answer

Nexus plans its own retrieval across the context’s curated knowledge, gathers the relevant evidence, and composes a single grounded answer with inline citations. Asking the company knowledge base “What is the refund policy?” returns something like:
Refund requests must be submitted within 30 days of purchase. After that window, customers can still get store credit, but not a direct refund. [1] [1] policies/refunds.md
The answer is grounded in your sources, and every claim cites the document it came from, so you can check it. To see how Nexus produced it, open the query’s trace. In the console, Nexus also suggests follow-up topics and saves the query as a session you can reopen from Sessions.

​
A multi-document answer

Questions that span multiple documents work the same way. Asking “Which purchases need approval, and what’s the threshold for each?” returns something like:
  • Standard expenses over $1,000 need manager approval. [1]
  • Vendor invoices over $10,000 need finance approval. [2]
  • Vendor invoices over $25,000 need VP and finance approval. [2]
[1] policies/expenses.pdf [2] policies/vendors.pdf
To answer this, Nexus draws on the artifacts it compiled from your policies during curation, gathering the matching rules from across your documents, each with its own citation. A plain RAG search returns only the passages closest to your question, so it can miss rules in other documents. Reaching across your whole corpus is a core reason to use Nexus over RAG. For exact counts and enumerations over structured data, define a SQLite artifact, covered in artifact formats. See how queries work for what the runtime does on each turn, or the overview for how Nexus compares to RAG.