Data Hub Overview
Turn your SQL, API, and S3 data into a queryable plane your models can reason over.
Data Hub connects your own data to your AI. It brings the records in your databases and APIs, and the documents in your S3 storage, into one place your custom models can query — so an assistant can answer questions across both your numbers and your files, and every answer points back to where it came from.
You work in three steps:
Your data → Your AI → Your answer
(connect) (attach) (ask)Every screen in Data Hub sits under one of those steps.
Architecture
You don't have to move or reshape your data by hand. Data Hub keeps a synced copy of your structured sources (databases and APIs) so an assistant can run queries against it without touching your production systems, and it turns your documents into searchable form. When you ask a question, the assistant decides on its own whether to look in the data, in the documents, or both.
For sources that support it, the synced copy can be kept near real time — a change in your database shows up in the copy within seconds, so an assistant always answers from current data.
Three Steps
Connect a data source
For structured data, open the Data Sources page and add a database (a read-only connection to PostgreSQL, MySQL, or SQL Server) or an API (its base URL, credentials, and the paths to pull). Turn on the sync toggle and run the first sync — the card shows the status right there: green when it's done, amber while it's running, red with a reason if it fails.
For documents, open a knowledge base and either upload files or sync a folder from S3. You pick the folder from a list — no path typing — and new, changed, or deleted files keep themselves in sync from then on.
Attach it to a model
In your custom model's data settings, tick the knowledge bases and data sources this assistant may use. For each database, choose how it should be queried: against the synced copy (recommended — keeps load off your production database) or the live source directly. Attaching a source is simply granting the assistant permission to use it.
Ask questions in natural language
In the Playground, or through the API, ask in plain language — for example, "What were the payment terms on our largest 2024 contract, and how much was it?" The assistant works out the answer for you: it can pull the figure from your data and find the terms in a contract PDF. Each answer carries source chips — the table it read from, and the full document path, like contracts/2024/acme-master-agreement.pdf — so you can trace it.
Because the assistant already knows the shape of your data, it picks the right table and writes accurate queries. Those table and column descriptions come from the Semantic Knowledge Graph. And when you ask for a chart, that same trusted answer becomes a dashboard card through Conversational BI.
For Admins
Everything lives under one entry point: admin panel → Data Platform, split into three areas.
Connection Pool
Data Hub Settings
Monitoring & Usage
Defaults and cadence. New connections inherit the default query target (synced copy or live source) and a sync cadence you set. Each tenant gets its own isolated storage, so one customer's data can never mix with another's.
Billing. Synced rows, queried rows, and document ingestion are metered and priced by plan — personal workspaces draw on the personal wallet, teams on the group wallet. Before billing is turned on, costs show as a dry-run estimate.
The admin inbox surfaces anything that needs attention — a failed sync, a document that didn't ingest, a connection with no assigned tenant — and each item links straight to the page that fixes it.