Connect your systems

A connector is the unit of integration. It represents one external system's interface: the agent reads and writes through it, and the platform handles authentication and request management.

Three ways to create one

Pick whichever matches the documentation you have. All three produce the same kind of connector, and everything downstream works the same way.

Import an OpenAPI spec (YAML, JSON, or a URL). The definition is parsed and every action is generated in one pass.
Create it through AI dialogue: describe the interface in plain language, or hand over a fragment of API documentation, and iterate on the generated action config.
Connect an MCP server. Model Context Protocol is supported natively, so third-party MCP servers work as agent tools directly.

Direct database access

PostgreSQL, MySQL, Oracle, SQL Server, and the domestic Chinese databases (Dameng, KingBase, GBase, HighGo) connect directly.

Each database connector generates three tools automatically (list_tables, describe_table, and query), so an agent can explore the schema before running a read-only query, with no manual tool configuration.

After it is connected

Connectors export as JSON templates you can share, and an existing connector can be forked and adapted to start a similar integration quickly.

Publish to an organization and teammates subscribe to reuse it, instead of each rebuilding the same integration.