The built-in connectors cover the systems everyone uses, but the systems that make your product yours are your own. The fastest way to give an agent access to them shouldn't be writing a tool adapter by hand for every endpoint.
Your API is already a tool catalog#
If you have an OpenAPI description of your service — and most teams do — you already have a machine-readable list of operations, their parameters, and their response shapes. Hand that spec to LoopLlama and each operation becomes a tool an agent can call, with arguments validated against the schema before any request goes out. For anything OpenAPI doesn't describe, a plain JSON schema works too.
Guardrails come along for free#
Because tools are declared, not improvised, the same controls apply to your endpoints as to the built-ins: scope what a workflow can reach, put destructive operations behind approval gates, and read back exactly which tool an agent called with which arguments in the run trace. Bringing your own tools doesn't mean bringing your own risk.