Tool use
The capability that lets a model call external functions, APIs, or services to retrieve information or take actions beyond generating text.
In more detail
Tool use is the bridge from a model that only produces text to a system that can act. Given a set of available tools, the model can decide to call one, supply the arguments, receive the result, and continue. That single capability is what lets a model look up a record, send a message, run a query, or update a system.
The reliability of tool use depends on more than the model. The tools must be well-described, their inputs validated, their failures handled, and their permissions constrained. A model that can call any tool freely is powerful and dangerous; a model whose tool access is scoped and audited is what a business can actually deploy.
Where this shows up at Ceven
Tool use is how Ceven's workflows touch the real world: steps call connectors across 1,000+ tools to read data and take actions. Ceven also exposes its own capabilities as tools through a hosted MCP server, so other AI systems can call them. In every case the consequential calls can require human approval and are recorded in the audit trail.