Integrations · on infrastructure you own
What your agents connect to,
and what leaves when they do.
Three kinds of connection: the sources an agent reads from, the tools it acts through and the model it calls. Each page says what is read, what is needed and what, if anything, leaves your environment.
The sources an agent reads from
Connectors
Documents, tickets, records and files are read into the Knowledge Engine on infrastructure you control, on a schedule you set.
The tools an agent acts through
MCP servers
Search the web, query a database, read a dashboard. Each row says who publishes the server and whether it sends data outside.
Where the model runs
Model providers
A model you serve on your own GPUs, one in your cloud account or a hosted one. The gateway holds the keys, the limits and the record.
What is the difference between a connector and an MCP server?
A connector reads a source into the Knowledge Engine, where it is indexed for agents to answer from. An MCP server gives an agent tools to call while it works, such as a web search or a database query.
Does connecting a source send our data to a public AI service?
No. A connector reads the source into the Knowledge Engine on infrastructure you own or control. What leaves depends on the tools and the model you choose, and each page says so.
Can we use a hosted model?
Yes. Then the prompt and the response go to that provider. The gateway holds the key, decides who may call which model and records the call. A model you serve yourself keeps everything inside.
Is this everything the platform connects to?
These pages list what the product's own lists name, and the MCP servers the use case pages name. The product's Catalog lists more MCP servers, and you can add your own.