Glossary

Tool calling

Tool calling is how an AI model acts outside its own text: it picks a tool, fills in the inputs and asks for the call. The tool might read an order, send an email or change a record.

In a run

Each tool call is one step in the run. Before it reaches the system, Ares checks it against the access policy and the credential for the job. The step shows as allowed, sent to a person or refused, and the record holds which.

Our word for it

The step: what the agent asked a system to do.

Where a wrong action happens

A model that only writes text can be wrong in a draft. A model that calls tools can be wrong in your ERP. That is why every tool call is checked, not only the first one in a job. A legitimate tool can still be asked for an action the job never needed, so the check covers the action, not only the tool.

The result is not the verdict

A tool call can fail while the model reports success. The outcome of a step should come from the result of the call, not from the model’s own account of it. The planner said done describes one case and the fix.

Where it stops

Ares checks whether a call is allowed and records the outcome. It does not check whether the data a tool returns is correct, and it does not make a model more accurate.

Run the sandbox to see tool calls allowed, sent to a person and refused.

Related terms

  • Model Context Protocol (MCP)The Model Context Protocol (MCP) is an open protocol for connecting AI agents to tools and data sources.
  • Tool cardOn Ares, a tool card is the published description of a tool server: its address, the tools it offers and the inputs each tool takes.

Tell us about one workflow.

We will say whether an agent should do it, and what the first step costs.

A senior member of the team replies within two business days.