Connect your data

Agents are only as useful as the data they can reach. CoreBase connects to four kinds of source, each with the egress posture your compliance needs. You manage them all from Panel → Sources.

​

Databases

​

Cloud-direct (paste a DSN)

The fastest path for a reachable database. In Sources → Add database, paste a connection string. CoreBase encrypts it per tenant (AES-GCM) and connects directly over TLS. Connections are read-only at the session level — the pool runs with default_transaction_read_only = on and a statement timeout. Mutating SQL is rejected by the database itself, not just by a parser. CoreBase auto-exposes four tools: list tables, describe a table (columns, primary key, foreign keys), fuzzy schema search, and run query (with result-size caps).

​

On-prem databases (via CoreMCP)

For SQL Server, Firebird, and anything behind a firewall, run the open-source CoreMCP agent inside your network. It holds the credentials, runs queries locally, and dials outbound only — no inbound ports. See Connect a database with CoreMCP.

​

REST & GraphQL APIs

In Sources → Add API, give CoreBase an OpenAPI/Swagger URL (REST) or a GraphQL endpoint. It introspects the spec and turns each operation into a typed tool the agent can call.

• REST — operations with an operationId become tools; path, query, and body parameters are typed from the spec.
• GraphQL — the schema is introspected; queries become parameterized tools.

APIs are read-only by default. Mutating operations (POST/PUT/PATCH/DELETE, GraphQL mutations) are not even exposed to the agent unless you enable writes on the connection — the guard is structural (we don’t produce the tool), not a prompt. See Governance.

​

Communications

​

SaaS apps

Connect 50+ apps (Jira, Gmail, GitHub, Notion, Stripe, HubSpot, Zendesk…) through managed OAuth — no per-integration code. See SaaS connectors.

​

Egress postures at a glance

See Zero raw data egress for the full breakdown.

​

Next