VantEdge
Platform/03 · Knowledge

Grounding and retrieval
over your private data.

Give agents accurate, permissioned access to the data they need — across databases and SaaS — without moving it or exposing credentials.

Context · Catalog
6 sources · 143 tables
Sources
postgres· Database127 tables
public
credit_notes“credit note”
orders“sales”
customers
salesforce· CRM9 objects
slack· Messaging4 channels
available: gmail drive office 365
Glossary · 12 terms
“credit note”→postgres.public.credit_notes
“last quarter”→temporal · current_qtr − 1
“APAC”→regions.name IN (JP, AU, KR, …)
credentials via K8s secret refs · never stored inline
01The problem

Agents guess without context

Without grounding in your real data, agents hallucinate confidently. But wiring every source securely — with the right schema, semantics, and access rules — is a project of its own.

02Capabilities

What this layer gives you.

First-class connectors

Postgres plus CData connectors for Salesforce, Gmail, Office 365, Drive, and Slack — and CSV/Excel.

Schema-aware grounding

Every source is cataloged with real schema and semantic labels — so generated queries match your tables and your business meaning, not a hallucinated shape.

Governed SQL pipeline

SQL is generated, reviewed, fixed, and validated by a governed pipeline before a single query runs.

Secret-reference credentials

Credentials resolve at connect time from env, file, or your secret manager — never stored inline.

03How it works
  1. 1Connect sourcesPoint at databases and SaaS tools using secret references, not copied credentials.
  2. 2Catalog & groundSchema and semantics are indexed so retrieval matches your business, not just string similarity.
  3. 3Query in contextAgents retrieve grounded, permissioned data through the deterministic execution layer.
Catalog/ workspace ·finance
143 tables · 12 terms
Sources
postgres127 tables
public
credit_notes“credit note”
orders“sales”
customers
salesforce9 objects
slack4 channels
available: gmail drive office 365
Glossary · 12 terms
“credit note”→postgres.public.credit_notes
“last quarter”→temporal · current_qtr − 1
“APAC”→regions.name IN (JP, AU, KR, …)
credentials via K8s secret refs · never stored inline

Grounded in your data

Answers from what's real — not what's likely.

04Where it sits

One integrated stack, GPU to Agent. This layer, in context.

00 · COMPUTEGPU Fleet01 · ACCESSAI Gateway02 · COMPUTEInference & Training Manager03 · KNOWLEDGEContext Enginedet · verified04 · TRUSTDeterministic Execution05 · BUILDAgent Builder
KNOWLEDGE
Connectors
Postgres · Salesforce · Gmail · O365 · Drive · Slack
Grounding
semantic catalog + glossary
SQL
generate · review · fix · validate
Secrets
referenced, never stored
in-placeno data movement·schemaaware grounding·0inline credentials
05The rest of the stack
AI GatewayOne governed door to every model.
Inference & Training ManagerYour GPU fleet, scheduled and served.
Deterministic ExecutionSame inputs, same outputs — provably.
Agent BuilderCompose, test, ship agents.
Platform overview