Skip to main content

Search and chat across company knowledge

Use VectorAmp to create a governed knowledge dataset that answers questions across company documents with citations back to the original source.

What you will build​

A searchable dataset that combines documents from implemented VectorAmp sources such as Google Drive, Confluence, S3, GCS, web pages, Jira, and file uploads. The result is a search and chat workflow where every useful answer can point back to the source document, page, row, or ticket.

Prerequisites​

  • A VectorAmp account and API key.
  • Access to the source systems you want to connect.
  • A clear owner for each source folder, space, bucket, or project.

Start with a few high-signal sources instead of dumping everything in at once.

SourceGood first useMetadata to keep
Google Driveshared docs, policies, PDFssource_system, folder, team, region
Confluenceproduct and process pagesspace, team, page_status
S3 or GCSexported docs, archives, reportsbucket, prefix, environment
Jiraissues, incidents, project notesproject, status, priority
File uploadcurated PDFs, DOCX, CSVsowner, document_type, reviewed

Step 1: Create a dataset​

Create a dataset in the VectorAmp dashboard or with the CLI. Dataset creation uses SABLE automatically.

npm install -g @vectoramp/cli
export VECTORAMP_API_KEY=vsk_...

vectoramp datasets create company-knowledge --metadata '{"owner":"knowledge-team"}'
vectoramp config use ds_123

Step 2: Connect implemented sources​

Use the dashboard connector flow, or create reusable sources with the CLI.

vectoramp sources gdrive google-drive-folder-id --name company-drive
vectoramp sources confluence https://acme.atlassian.net/wiki --name product-confluence \
  --config '{"spaces":["PROD"],"include_attachments":true}'
vectoramp sources s3 s3://acme-knowledge/policies --name policy-archive \
  --config '{"sync_mode":"incremental"}'

Step 3: Run ingestion​

Start ingestion into the dataset and monitor the job.

vectoramp --dataset ds_123 sources ingest s3 s3://acme-knowledge/policies \
  --config '{"sync_mode":"incremental"}'

vectoramp jobs list --dataset-id ds_123
vectoramp jobs get job_123 --poll --interval 5000 --timeout 600000

Step 4: Ask cited questions​

In the Intelligence UI, ask questions that require source evidence:

  • “What changed in the 2026 renewal policy?”
  • “Which contracts mention EU data residency?”
  • “Summarize open implementation risks from platform docs.”

Use citations to verify each answer before forwarding it to another team.

Step 5: Search with filters​

Filters keep broad company search from becoming a generic chatbot.

vectoramp --dataset ds_123 datasets search \
  "Which policies mention EU data residency?" \
  --filter region=eu \
  --filter document_type=policy \
  --top-k 10

Python SDK equivalent:

from vectoramp import VectorAmp

client = VectorAmp(api_key="vsk_...")
dataset = client.datasets.get("ds_123")

results = dataset.search(
    text="Which policies mention EU data residency?",
    top_k=10,
    filters={"region": "eu", "document_type": "policy"},
    include_documents=True,
)

Validation checklist​

  • Important documents appear in dataset document listing.
  • The first answers include citations to expected sources.
  • Source owners can identify stale or incorrect files.
  • Filters produce narrower, more relevant results.

Production notes​

  • Use separate datasets when data ownership or access policies differ significantly.
  • Keep metadata values consistent. Avoid both EU and eu for the same concept.
  • Review ingestion failures weekly.
  • Prefer cited answers in workflows where people will make decisions from the output.