The entrances

These entrances and no others:

import { defineAgent, defineJob, defineConfig } from "@chloejs/core";
import { deliverEmail, readEmailMessages } from "@chloejs/core/services"; // the work, for a job
import * as gmail from "@chloejs/core/tools/gmail";                     // a tool to bind
import { telegramChannel } from "@chloejs/core/channels";               // a channel to bind
import { mcpConnection } from "@chloejs/core/connections";              // a service's MCP server
import { calls, expectations } from "@chloejs/core/scorers";            // marking a run
import { every } from "@chloejs/core/timer";                            // when a job runs
import { about, is } from "@chloejs/core/test";                         // testing a job

Each set of tools is its own entrance: @chloejs/core/tools/gmail, /calendar, /drive, /resend, /email, /web and /fs. Import one whole, import * as gmail, and each function is a verb and a noun: gmail.readEmail(). The name the model sees is the key you give it, by convention the two together: { gmailReadEmail: gmail.readEmail({ ... }) }. A job does the same work through @chloejs/core/services, whose functions are named for the work, like deliverEmail(), so a job and a model never reach for the same name.

A tool of your own is made with the AI SDK's tool(), from "ai".

Never import a file inside the package by its path: it breaks the day a folder moves. What is not on these lists may change without notice. The reference is generated from them, so it is exactly what is published.