How to Stream AI Replies Live in Discord by Editing Messages

Make Discord answers feel like ChatGPT by streaming tokens and editing the message as they arrive.

9 minAdvanced

Waiting ten seconds for a wall of text feels broken. The fix is to stream: post a placeholder, then edit the message as tokens arrive so the answer appears to type itself. This guide shows the pattern with a throttle so you do not hit Discord's rate limits.

What you need

A working discord.js bot (see the auto-reply guide)
An AI provider that supports streaming responses
Node 18+ installed

Do not edit on every token

Editing a message on each token will blow past Discord's edit rate limit in seconds. Batch updates on a timer, roughly every 800 to 1200 milliseconds.

Step 1: Open a streaming request

Set stream: true on the chat completion. Instead of one response you get an async iterable of chunks, each carrying a small delta of text.

stream.js

const stream = await ai.chat.completions.create({
  model: "gpt-5-mini",
  stream: true,
  messages: [{ role: "user", content: prompt }],
});

Step 2: Post a placeholder and edit on a timer

Send a single character first to get a message you can edit. Accumulate the streamed text in a buffer and flush it to the message on an interval. Clear the interval and do a final edit when the stream ends.

stream.js (continued)

const sent = await msg.reply("...");
let buffer = "";
let dirty = false;

const timer = setInterval(async () => {
  if (!dirty) return;
  dirty = false;
  await sent.edit(buffer.slice(0, 1990) || "...");
}, 1000);

for await (const chunk of stream) {
  const delta = chunk.choices[0]?.delta?.content || "";
  if (delta) {
    buffer += delta;
    dirty = true;
  }
}

clearInterval(timer);
await sent.edit(buffer.slice(0, 1990));

#chat - mid stream

You

@MyHelper write a haiku about deadlines

Agent

The clock leans closer

Agent

The clock leans closer / coffee cools beside the screen

The reply grows in place every second.

Add a cursor for polish

Append a small marker like a block character to the buffer while streaming and strip it on the final edit. It signals to readers that more text is coming.

Step 3: Run and watch it type

zsh - discord-stream

$node index.js

Logged in as MyHelper#4821

Mention the bot and watch the reply build

Result

Long answers now feel responsive because users see text appear within a second and watch it grow. The timer keeps you safely under Discord's rate limits even for paragraph-length replies.