Tool call

A tool call is a request from a large language model (LLM) for the application to run a named function, such as a search or an API call, with arguments the model chose. The application runs the function and sends the result back to the model. The capability is also called function calling.

How it works

The application sends the model a list of tools, each with a name, a description and a JSON schema for its arguments. The model replies with either a final answer or a tool call, marked by its finish reason. In a triage agent, the model might ask for check_service_health("catalog"), read the result, then ask for check_dependencies. Each one appears as an execute_tool span in OpenTelemetry.

Why it matters in production

Tool calls reach systems you may not own, so they add latency, failures and security exposure to every request. Count tool calls per request: a rising count means the agent is doing more work per answer, which costs tokens even when nothing fails.

See also

• Agent observability

• Prompt injection

• AI observability

• Token (LLM)

On the One2N blog

• AI agent observability: tool calls, turns and runaway loops

• What is AI observability? A guide for LLM and agent apps

• How to evaluate AI SRE agents for production?

References

1. Origin of the term

2. Model Context Protocol specification

3. Toolformer: language models can teach themselves to use tools (Schick et al., 2023)