Typed sections
Your source answers in projectable sections, the same as a built-in provider. Ask for the fields you read; only those cross the wire.
Data / Bring your own
Wrap any source. It behaves like the rest.
The 53-provider catalogue is a starting point, not a fence. If the source your
agent needs is not in it, wrap it — REST, GraphQL or an MCP server — and it becomes a
first-class connector that answers exactly like the built-ins, so your agent learns nothing
new to use it.
RESTGraphQLMCP server → one interface
Earned freshness
A custom source gets the same freshness discipline — a long cache TTL only where it can be justified, and a safe default where it cannot.
First-class in discovery
It appears in your catalogue and answers discovery calls like any other provider, so an agent finds and uses it without knowing it was one you added.
The same 454 live endpoints, plus whatever you add — all reached the one way. Custom connectors are configured in the console.
Point one agent at it.
Connect over MCP or REST. If your agents read more than one source, the context bill is the first thing you will see move.
Measured on our own agent-swarm workload. swarmsDB is battle-tested, not yet widely adopted — external traction is the next milestone.