Foundations: what an AI agent actually is

const BASE = process.env.OTT_API_BASE
  ?? "https://ontimetrucking.com/api/v1";

// 1. What the model reads: a name, a description, a schema.
export const getQuoteTool = {
  name: "get_quote",
  description: "Price an LTL shipment between two US ZIPs. " +
    "Returns customer-facing all-in prices.",
  input_schema: {
    type: "object",
    required: ["originZip", "destZip", "weightLbs"],
    properties: {
      originZip: { type: "string", pattern: "^[0-9]{5}$" },
      destZip: { type: "string", pattern: "^[0-9]{5}$" },
      weightLbs: { type: "number", minimum: 1, maximum: 30000 },
    },
  },
};

type Args = { originZip: string; destZip: string; weightLbs: number };

// 2. What runs when the model calls it: a plain function.
export async function getQuote(a: Args) {
  const res = await fetch(BASE + "/quotes", {
    method: "POST",
    headers: {
      Authorization: "Bearer " + process.env.OTT_API_KEY,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      originZip: a.originZip,
      destZip: a.destZip,
      commodities: [{
        freightClass: "100",
        weightLbs: a.weightLbs,
        pieces: 1,
        dimsIn: { length: 48, width: 40, height: 48 },
      }],
    }),
  });
  // Return errors as data so the model can read them.
  if (!res.ok) {
    return { error: res.status, detail: await res.text() };
  }
  const q = await res.json();
  return {
    quoteId: q.quoteId,
    options: q.options,
    expiresAt: q.expiresAt,
  };
}

// 3. Try it without a model: call the tool the way a model would.
const args = { originZip: "11735", destZip: "10001", weightLbs: 500 };
const result = await getQuote(args);
console.log(JSON.stringify(result, null, 2));

const BASE = process.env.OTT_API_BASE
  ?? "https://ontimetrucking.com/api/v1";

// 1. What the model reads: a name, a description, a schema.
export const getQuoteTool = {
  name: "get_quote",
  description: "Price an LTL shipment between two US ZIPs. " +
    "Returns customer-facing all-in prices.",
  input_schema: {
    type: "object",
    required: ["originZip", "destZip", "weightLbs"],
    properties: {
      originZip: { type: "string", pattern: "^[0-9]{5}$" },
      destZip: { type: "string", pattern: "^[0-9]{5}$" },
      weightLbs: { type: "number", minimum: 1, maximum: 30000 },
    },
  },
};

type Args = { originZip: string; destZip: string; weightLbs: number };

// 2. What runs when the model calls it: a plain function.
export async function getQuote(a: Args) {
  const res = await fetch(BASE + "/quotes", {
    method: "POST",
    headers: {
      Authorization: "Bearer " + process.env.OTT_API_KEY,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      originZip: a.originZip,
      destZip: a.destZip,
      commodities: [{
        freightClass: "100",
        weightLbs: a.weightLbs,
        pieces: 1,
        dimsIn: { length: 48, width: 40, height: 48 },
      }],
    }),
  });
  // Return errors as data so the model can read them.
  if (!res.ok) {
    return { error: res.status, detail: await res.text() };
  }
  const q = await res.json();
  return {
    quoteId: q.quoteId,
    options: q.options,
    expiresAt: q.expiresAt,
  };
}

// 3. Try it without a model: call the tool the way a model would.
const args = { originZip: "11735", destZip: "10001", weightLbs: 500 };
const result = await getQuote(args);
console.log(JSON.stringify(result, null, 2));