v2 is a from-scratch redesign, no v1 compatibility. Coming from v1? See the v1 -> v2 migration guide in the changelog.
npm install node-telegram-bot-api@next**Runs on Bun, modern Node.js, Deno, Cloudflare Workers and Vercel Functions λ
import { Bot, InlineKeyboardBuilder } from "node-telegram-bot-api";
import { run } from "node-telegram-bot-api/node"; // managed runner: wires Ctrl-C to bot.stop()
const bot = new Bot(process.env.BOT_TOKEN!);
// commands, regex and update types are all middleware - registration order wins
bot.command("start", (ctx) => ctx.reply("Hi! Send me anything."));
bot.hears(/echo (.+)/, (ctx) => ctx.reply(ctx.match![1]!));
bot.on("message", (ctx) =>
ctx.reply("Pick one:", {
reply_markup: new InlineKeyboardBuilder()
.text("👍", "up")
.text("👎", "down")
.build(),
}),
);
// 🔘 a tapped inline button comes back as a callback_query
bot.on("callback_query", async (ctx) => {
await ctx.answerCallbackQuery({ text: `You tapped ${ctx.callbackQuery!.data}` });
});
await run(bot); // core-only alternative that runs anywhere: await bot.startPolling()Api mirrors the wire API 1:1 - one method per Bot API method, each taking a single params object.
import { Api } from "node-telegram-bot-api";
const api = new Api(process.env.BOT_TOKEN!);
const me = await api.getMe();
await api.sendMessage({ chat_id: 12345, text: "hello" });
// the same client is also on bot.api and ctx.apikoa-style middleware around every update; on/command/hears are filters in the same chain. Wrap downstream work with await next().
// ⏱️ time every update - and catch anything thrown downstream
bot.use(async (ctx, next) => {
const start = Date.now();
try {
await next();
} finally {
console.log(`update took ${Date.now() - start}ms`);
}
});
// 🧯 last-resort error handler
bot.catch((err, ctx) => console.error("handler failed", err));Structured fields are plain typed objects - pass a literal or use a fluent builder; the pipeline serializes either.
import { Bot, InlineKeyboardBuilder, ReplyKeyboardBuilder, EntityBuilder } from "node-telegram-bot-api";
const bot = new Bot(process.env.BOT_TOKEN!);
// 🎛️ inline keyboard as reply_markup
await bot.api.sendMessage({
chat_id,
text: "Choose:",
reply_markup: new InlineKeyboardBuilder()
.text("A", "a")
.url("Docs", "https://core.telegram.org/bots/api")
.row()
.text("B", "b")
.build(),
});
// ⌨️ reply keyboard as reply_markup
await bot.api.sendMessage({
chat_id,
text: "Yes or no?",
reply_markup: new ReplyKeyboardBuilder()
.text("Yes")
.text("No")
.build({ resize_keyboard: true }),
});
// ✍️ rich text - EntityBuilder computes UTF-16 offsets for you
const { text, entities } = new EntityBuilder()
.plain("Hello ")
.bold("world")
.link("docs", "https://github.com/yagop/node-telegram-bot-api")
.build();
await bot.api.sendMessage({ chat_id, text, entities });
// any structured field is just a plain object - no wrapper needed
await bot.api.sendMessage({ chat_id, text: "hi", link_preview_options: { is_disabled: true } });A bare string is always a file_id or URL. Wrap raw bytes to upload them. Pass a
filename with the right extension - the core does no content sniffing, so the name is
what Telegram sees (fromPath uses the basename).
import { Bot, InputFile, MediaGroupBuilder } from "node-telegram-bot-api";
import { fromPath } from "node-telegram-bot-api/node";
const bot = new Bot(process.env.BOT_TOKEN!);
// upload from disk (Node only)
await bot.api.sendPhoto({ chat_id, photo: await fromPath("./cat.jpg") });
// upload raw bytes (web-standard, runs anywhere)
await bot.api.sendDocument({ chat_id, document: new InputFile(bytes, { filename: "report.pdf" }) });
// a raw InputFile nested in a structure is auto-hoisted to an attach:// part
await bot.api.sendMediaGroup({
chat_id,
media: [
{ type: "photo", media: new InputFile(bytesA, { filename: "a.jpg" }), caption: "A" },
{ type: "photo", media: "https://telegram.org/example/photo.jpg" },
],
});
// MediaGroupBuilder: optional sugar for the same array
await bot.api.sendMediaGroup({
chat_id,
media: new MediaGroupBuilder()
.photo({ media: new InputFile(bytesA, { filename: "a.jpg" }), caption: "A" })
.photo({ media: "https://telegram.org/example/photo.jpg" })
.build(),
});Builders cover the other attach:// methods; each .build() returns the plain shape.
import {
Bot,
InputFile,
StickerSetBuilder,
StaticProfilePhotoBuilder,
PhotoStoryBuilder,
} from "node-telegram-bot-api";
const bot = new Bot(process.env.BOT_TOKEN!);
// collect a sticker set
await bot.api.createNewStickerSet({
user_id,
name,
title,
stickers: new StickerSetBuilder()
.add({ sticker: new InputFile(pngBytes, { filename: "sticker.png" }), format: "static", emoji_list: ["🙂"] })
.build(),
});
// a single sticker is a plain InputSticker - no builder needed
await bot.api.addStickerToSet({
user_id,
name,
sticker: { sticker: new InputFile(pngBytes, { filename: "sticker.png" }), format: "static", emoji_list: ["🙂"] },
});
// profile photo: Static / AnimatedProfilePhotoBuilder
await bot.api.setMyProfilePhoto({
photo: new StaticProfilePhotoBuilder({ photo: new InputFile(pngBytes, { filename: "avatar.png" }) }).build(),
});
// story: Photo / VideoStoryBuilder
await bot.api.postStory({
business_connection_id,
active_period,
content: new PhotoStoryBuilder({ photo: new InputFile(pngBytes, { filename: "story.png" }) }).build(),
});The web-standard callback is a pure (Request) => Promise<Response> - one function for every serverless runtime.
Cloudflare Workers / Bun.serve / Deno Deploy / Vercel Edge:
import { Bot, webhookCallback } from "node-telegram-bot-api";
const bot = new Bot(TOKEN);
bot.on("message", (ctx) => ctx.reply("hi from the edge"));
export default {
fetch: webhookCallback(bot, { secretToken: SECRET }),
};By default the callback awaits your handler before 200. For slow handlers, opt into early-ACK:
export default {
// ✅ return 200 immediately, then finish the handler in the background
// waitUntil keeps the platform alive until it settles (fastAck: true = fire-and-forget)
fetch: (req: Request, _env: unknown, ctx: { waitUntil(promise: Promise<unknown>): void }) =>
webhookCallback(bot, { secretToken: SECRET, waitUntil: (p) => ctx.waitUntil(p) })(req),
};Next.js App Router (app/api/bot/route.ts):
import { Bot, nextAppWebhook } from "node-telegram-bot-api";
const bot = new Bot(process.env.BOT_TOKEN!);
export const POST = nextAppWebhook(bot, { secretToken: process.env.SECRET });Express (mount on an app you already have):
import express from "express";
import { Bot, registerExpressWebhook } from "node-telegram-bot-api";
const app = express();
const bot = new Bot(TOKEN);
registerExpressWebhook(bot, app, { path: "/telegram", secretToken: SECRET });
app.listen(3000);Self-hosted Node server (node-telegram-bot-api/node):
import { Bot } from "node-telegram-bot-api";
import { createWebhookServer, startWebhook } from "node-telegram-bot-api/node";
// Low-level: you own the server and the port.
const server = createWebhookServer(new Bot(TOKEN), { path: "/telegram", secretToken: SECRET });
server.listen(8080);
// Or the managed one-liner (listen + graceful shutdown, the webhook peer of run()):
await startWebhook(new Bot(TOKEN), { port: 8080, path: "/telegram", secretToken: SECRET });Register the URL once: api.setWebhook({ url, secret_token }). The secret_token is the only thing authenticating callers (payloads aren't signed) - treat it as required in production, and terminate TLS at your proxy.
Errors expose structured fields, so you branch on values, not message text.
import { TelegramApiError, NetworkError, TimeoutError } from "node-telegram-bot-api";
try {
await api.sendMessage({ chat_id, text });
} catch (err) {
// 🔁 429s are auto-retried (honoring retry_after) by default - this is the manual form
if (err instanceof TelegramApiError && err.errorCode === 429) {
await sleep((err.retryAfter ?? 1) * 1000);
} else if (err instanceof NetworkError || err instanceof TimeoutError) {
// transient transport failure
}
}An error thrown by a handler never stops the bot. It is routed to the error boundary, which by default logs it via console.error and consumes the update - polling keeps pumping, and a webhook delivery is ACKed (Telegram does not redeliver it). Install your own boundary with bot.catch():
bot.catch((err, ctx) => {
console.error("update", ctx.update.update_id, "failed:", err);
});Throwing from the boundary opts back into fail-loud: startPolling() rejects (the update was never confirmed, so Telegram redelivers it on restart) and webhookCallback responds 500 (Telegram redelivers). bot.catch((err) => { throw err; }) is the explicit fail-loud opt-in.
Safe defaults out of the box - set only what you want to change.
import { Api } from "node-telegram-bot-api";
const api = new Api(TOKEN, {
// 🔁 retries 429 (retry_after first), network/timeout/5xx with jittered backoff
maxRetries: 2, // default 2
retryBackoffMs: 300, // default 300
// 🚦 opt-in throttle (requests/sec); omit for zero overhead
rateLimit: { global: 30, perChat: 1 },
});Long polling resumes through transient errors instead of dying on the first blip:
import { longPoll } from "node-telegram-bot-api";
for await (const update of longPoll(api, {
timeout: 30,
retry: true, // default true - resume on transient errors, keep the offset
maxBackoffMs: 60_000, // default 60s - cap between failed polls
onError: (err) => console.warn("poll failed, backing off", err),
}, signal)) {
// ... fatal 4xx still stops the loop; an aborted signal returns cleanly
}longPoll is a plain async generator - for await, take(n), filter, batch or fan out as you like.
import { Api, longPoll } from "node-telegram-bot-api";
const api = new Api(TOKEN);
const ac = new AbortController();
for await (const update of longPoll(api, { timeout: 30 }, ac.signal)) {
console.log(update.update_id);
}Set DEBUG (the debug convention) to trace request lifecycle, polling and webhooks to stderr:
DEBUG="node-telegram-bot-api:*" node app.js
# node-telegram-bot-api:transport -> sendMessage
# node-telegram-bot-api:transport <- sendMessage ok +142msNamespaces: :transport, :polling, :webhook (filter, or exclude one with a leading -). Tracing is Node-only - wired up by importing node-telegram-bot-api/node; on edge runtimes it's an inert no-op.
bun run generate:types # regenerate types + client from the live Bot API docs
bun run check # tsc (strict) + core-isolation lint + edge bundle + unit tests
bun run build # emit dist/MIT