Agents

Built-in Tools

Built-in tools are executable functions that run server-side. Enable them on any agent to give it capabilities like web search, math, code execution, and more — no setup required.

Enabling tools

In the agent editor, go to the Skills tab and toggle on the built-in tools you want. They appear as OpenAI function calls in the agent's tool list. The orchestrator executes them automatically when the LLM invokes them.

Available Tools

Web Search

builtin_web_search

Calculator

builtin_calculator

Code Executor

builtin_code_executor

JSON Formatter

builtin_json_formatter

Text Analyzer

builtin_text_analyzer

Web Search

Search the web for information or read the content of a specific URL. Combines search and page reading into a single tool.

Tool ID

builtin_web_search

Function Name

web_search

Parameters

query

Search query string. Returns top results from the web.

string, optional

url

Specific URL to read and extract content from.

string, optional

max_results

Maximum number of search results to return (default: 5).

number, optional

max_length

Maximum character length of extracted page content (default: 5000).

number, optional

Either query or url must be provided. Internal/private IPs are blocked for security.

Calculator

Evaluate mathematical expressions with support for arithmetic, functions, and constants. Useful for agents that need precise calculations instead of LLM approximations.

Tool ID

builtin_calculator

Parameters

expression

Mathematical expression to evaluate.

string, required

Supported Operations

+ - * / % — Arithmetic

^ — Exponentiation

sqrt, abs, floor, ceil, round

sin, cos, tan, log

min(a,b), max(a,b)

PI, E — Constants

javascript

// Example calls the agent might make:
calculator({ expression: "sqrt(144) + 2^3" })     // → 20
calculator({ expression: "sin(PI/2) * 100" })      // → 100
calculator({ expression: "min(42, 17, 89)" })      // → 17

Code Executor

Execute JavaScript code in a sandboxed environment. Supports basic operations, JSON processing, array methods, and Math. External access (require, fetch, eval) is blocked for security.

Tool ID

builtin_code_executor

Parameters

code

JavaScript code to execute.

string, required

input

Data passed to the code as the input variable.

any, optional

Security

The sandbox blocks require(), eval(),process, fetch, global,import(), async/await, and__proto__ access. Execution times out after a configurable limit.

JSON Formatter

Validate, format, minify, and query JSON data. Supports JSONPath-like queries for extracting nested values.

Tool ID

builtin_json_formatter

Parameters

json

JSON string to process.

string, required

operation

Operation to perform: validate, prettify,minify, query, keys, or values.

string, default: "validate"

path

Dot-notation path for query operation (e.g. data.users[0].name).

string, optional

Text Analyzer

Analyze text for word count, character count, readability scores, reading time, vocabulary richness, and word frequency.

Tool ID

builtin_text_analyzer

Parameters

text

Text to analyze.

string, required

include_readability

Include readability scores (Flesch-Kincaid, etc.). Default: true.

boolean, optional

top_words

Number of most frequent words to return. Default: 0 (disabled).

number, optional

Output includes

Word count & character count

Sentence & paragraph count

Reading & speaking time

Average word length

Vocabulary richness

Readability scores

Enabling Tools in Agent Config

json

{
  "builtinTools": [
    "builtin_web_search",
    "builtin_calculator",
    "builtin_code_executor"
  ]
}

The orchestrator converts these IDs into OpenAI-compatible function definitions and adds them to the LLM request. When the model calls a tool, the orchestrator executes it server-side and returns the result in the conversation.

Next: Custom Tools & Webhooks

Define your own tools with webhook endpoints.