miatzSign in

Letting a model call your internal systems through a tool interface

The model is handed a typed catalogue of callable functions — look up an order, check stock in a warehouse, raise a ticket — and chooses which one to invoke and with what arguments. A separate layer decides whether that invocation is permitted: reads run unattended, anything that writes waits behind an allow-list or a human confirmation.

Effort
Weeks of work
Skill level
Some technical skill
Organisation size
Mid-market
Value
Time saved, Quality

Tools named for this

  • A function-calling interface where each tool is declared with a typed schema
  • A protocol server that exposes an existing system's operations as callable tools
  • A permission layer that decides which calls may execute unattended and which need a person

What to check before you ship it in India

  • A tool that reads or writes a customer record is processing personal data whether a person or a model picked it. Section 8(1) makes the fiduciary responsible for compliance in respect of any processing undertaken by it or on its behalf by a Data Processor, irrespective of any agreement to the contrary — so handing the choice of call to a model delegates the decision and none of the duty.
  • The tool catalogue is an execution surface. The protocol specification is explicit that tools represent arbitrary code execution, that a tool's own description of its behaviour should be treated as untrusted unless it comes from a trusted server, and that the protocol cannot enforce any of this at the protocol level — so a catalogue assembled from third-party servers needs the same review as a dependency.

Sources

Every claim on this page traces to one of these, on the date it was read.