Tool Invocation

Evaluate whether LLM tool calls have correct arguments and formatting

Overview

The Tool Invocation evaluator determines whether an LLM invoked a tool correctly with proper arguments, formatting, and safe content. This evaluator focuses on the how of tool calling - validating that the invocation itself is well-formed - rather than whether the right tool was selected.

When to Use

Use the Tool Invocation evaluator when you need to:

• Validate tool call arguments - Ensure all required parameters are present with correct values
• Check JSON formatting - Verify tool calls are properly structured
• Detect hallucinated fields - Identify when the LLM invents parameters not in the schema
• Audit for unsafe content - Check that arguments don’t contain PII or sensitive data
• Evaluate multi-tool invocations - Validate when the LLM calls multiple tools at once

This evaluator validates tool invocation correctness, not tool selection. For evaluating whether the right tool was chosen, use the Tool Selection evaluator instead.

Supported Levels

The level of an evaluator determines the scope of the evaluation in OpenTelemetry terms. Some evaluations are applicable to individual spans, some to full traces or sessions, and some are applicable at multiple levels.

Level	Supported	Notes
Span	Yes	For LLM spans that contain tool calls. Evaluate individual tool-calling decisions.

Relevant span kinds: Tool spans or LLM spans with tool calls, particularly in agentic applications.

Input Requirements

The Tool Invocation evaluator requires three inputs:

Field	Type	Description
input	string	The conversation context (can include multi-turn history)
available_tools	string	Tool schemas (JSON schema or human-readable format)
tool_selection	string	The LLM’s tool invocation(s) with arguments

In TypeScript, the fields use camelCase: availableTools and toolSelection.

Formatting Tips

While you can pass full JSON representations for each field, human-readable formats typically produce more accurate evaluations.

input (conversation context adapted from input messages):

User: I need to book a flight from New York to Los Angeles
Assistant: I'd be happy to help you book a flight. When would you like to travel?
User: Tomorrow morning, the earliest available

available_tools (tool descriptions adapted by JSON schemas):

book_flight: Book a flight between two cities
  - origin (required): Departure city code (e.g., "JFK", "LAX")
  - destination (required): Arrival city code
  - date (required): Flight date in YYYY-MM-DD format
  - time_preference (optional): "morning", "afternoon", or "evening"

search_hotels: Search for hotel accommodations
  - city (required): City name or code
  - check_in (required): Check-in date in YYYY-MM-DD format
  - check_out (required): Check-out date in YYYY-MM-DD format

tool_selection (the LLM’s tool invocation adapted from tool_calls in the output):

book_flight(origin="JFK", destination="LAX", date="2024-01-15", time_preference="morning")

Additional tips:

• Include full conversation context - The evaluator considers the entire conversation history to validate argument values
• Multi-tool invocations are supported - If the LLM calls multiple tools, include all invocations in the tool_selection field

Output Interpretation

The evaluator returns a Score object with the following properties:

Property	Value	Description
label	"correct" or "incorrect"	Classification result
score	1.0 or 0.0	Numeric score (1.0 = correct, 0.0 = incorrect)
explanation	string	LLM-generated reasoning for the classification
direction	"maximize"	Higher scores are better
metadata	object	Additional information such as the model name. When tracing is enabled, includes the trace_id for the evaluation.

Usage Examples

Python

from phoenix.evals import LLM
from phoenix.evals.metrics import ToolInvocationEvaluator

# Initialize the LLM client
llm = LLM(provider="openai", model="gpt-4o")

# Create the evaluator
tool_invocation_eval = ToolInvocationEvaluator(llm=llm)

# Inspect the evaluator's requirements
print(tool_invocation_eval.describe())

# Evaluate a tool invocation using human-readable format
eval_input = {
    "input": """User: I need to book a flight from New York to Los Angeles
Assistant: I'd be happy to help you book a flight. When would you like to travel?
User: Tomorrow morning, the earliest available""",
    "available_tools": """book_flight: Book a flight between two cities
- origin (required): Departure city code (e.g., "JFK", "LAX")
- destination (required): Arrival city code
- date (required): Flight date in YYYY-MM-DD format
- time_preference (optional): "morning", "afternoon", or "evening"

search_hotels: Search for hotel accommodations
- city (required): City name or code
- check_in (required): Check-in date in YYYY-MM-DD format
- check_out (required): Check-out date in YYYY-MM-DD format""",
    "tool_selection": 'book_flight(origin="JFK", destination="LAX", date="2024-01-15", time_preference="morning")'
}

scores = tool_invocation_eval.evaluate(eval_input)
print(scores[0])
# Score(name='tool_invocation', score=1.0, label='correct', ...)

TypeScript

import { createToolInvocationEvaluator } from "@arizeai/phoenix-evals";
import { openai } from "@ai-sdk/openai";

// Create the evaluator
const toolInvocationEvaluator = createToolInvocationEvaluator({
  model: openai("gpt-4o"),
});

// Evaluate a tool invocation using human-readable format
const result = await toolInvocationEvaluator.evaluate({
  input: `User: I need to book a flight from New York to Los Angeles
Assistant: I'd be happy to help you book a flight. When would you like to travel?
User: Tomorrow morning, the earliest available`,
  availableTools: `book_flight: Book a flight between two cities
- origin (required): Departure city code (e.g., "JFK", "LAX")
- destination (required): Arrival city code
- date (required): Flight date in YYYY-MM-DD format
- time_preference (optional): "morning", "afternoon", or "evening"

search_hotels: Search for hotel accommodations
- city (required): City name or code
- check_in (required): Check-in date in YYYY-MM-DD format
- check_out (required): Check-out date in YYYY-MM-DD format`,
  toolSelection: 'book_flight(origin="JFK", destination="LAX", date="2024-01-15", time_preference="morning")',
});

console.log(result);
// { score: 1, label: "correct", explanation: "..." }

Using Input Mapping

When your data has different field names, use input mapping.

Python

from phoenix.evals import LLM
from phoenix.evals.metrics import ToolInvocationEvaluator

llm = LLM(provider="openai", model="gpt-4o")
tool_invocation_eval = ToolInvocationEvaluator(llm=llm)

eval_input = {
    "conversation": """User: Search for hotels in Paris
Assistant: I can help you find hotels. What are your check-in and check-out dates?
User: March 15th to March 20th""",
    "tools_schema": """search_hotels: Search for hotel accommodations
- city (required): City name or code
- check_in (required): Check-in date in YYYY-MM-DD format
- check_out (required): Check-out date in YYYY-MM-DD format""",
    "llm_tool_call": 'search_hotels(city="Paris", check_in="2024-03-15", check_out="2024-03-20")'
}

input_mapping = {
    "input": "conversation",
    "available_tools": "tools_schema",
    "tool_selection": "llm_tool_call"
}

scores = tool_invocation_eval.evaluate(eval_input, input_mapping)

For more details on input mapping options, see Input Mapping.

TypeScript

import { bindEvaluator, createToolInvocationEvaluator } from "@arizeai/phoenix-evals";
import { openai } from "@ai-sdk/openai";

const toolInvocationEvaluator = createToolInvocationEvaluator({
  model: openai("gpt-4o"),
});

const boundEvaluator = bindEvaluator(toolInvocationEvaluator, {
  inputMapping: {
    input: "conversation",
    availableTools: "toolsSchema",
    toolSelection: "llmToolCall",
  },
});

const result = await boundEvaluator.evaluate({
  conversation: `User: Search for hotels in Paris
Assistant: I can help you find hotels. What are your check-in and check-out dates?
User: March 15th to March 20th`,
  toolsSchema: `search_hotels: Search for hotel accommodations
- city (required): City name or code
- check_in (required): Check-in date in YYYY-MM-DD format
- check_out (required): Check-out date in YYYY-MM-DD format`,
  llmToolCall: 'search_hotels(city="Paris", check_in="2024-03-15", check_out="2024-03-20")',
});

For more details on input mapping options, see Input Mapping.

Configuration

For LLM client configuration options, see Configuring the LLM.

Viewing and Modifying the Prompt

You can view the latest versions of our prompt templates on GitHub. The evaluators are designed to work well in a variety of contexts, but we highly recommend modifying the prompt to be more specific to your use case. Feel free to adapt them.

Python

from phoenix.evals.metrics import ToolInvocationEvaluator
from phoenix.evals import LLM, ClassificationEvaluator

llm = LLM(provider="openai", model="gpt-4o")
evaluator = ToolInvocationEvaluator(llm=llm)

# View the prompt template
print(evaluator.prompt_template)

# Create a custom evaluator based on the built-in template
custom_evaluator = ClassificationEvaluator(
    name="tool_invocation",
    prompt_template=evaluator.prompt_template,  # Modify as needed
    llm=llm,
    choices={"correct": 1.0, "incorrect": 0.0},
    direction="maximize",
)

TypeScript

import { TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG, createToolInvocationEvaluator } from "@arizeai/phoenix-evals";
import { openai } from "@ai-sdk/openai";

// View the prompt template
console.log(TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG.template);

// Create a custom evaluator with a modified template
const customEvaluator = createToolInvocationEvaluator({
  model: openai("gpt-4o"),
  promptTemplate: TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG.template, // Modify as needed
});

Using with Phoenix

Evaluating Traces

Run evaluations on traces collected in Phoenix and log results as annotations:

• Evaluating Phoenix Traces
• Logging LLM Evaluations

Running Experiments

Use the Tool Invocation evaluator in Phoenix experiments:

• Using Evaluators in Experiments

API Reference

• Python: ToolInvocationEvaluator
• TypeScript: createToolInvocationEvaluator

Related

• Tool Selection Evaluator - For evaluating whether the right tool was chosen
• Correctness Evaluator - For evaluating factual accuracy