Bot Reply Delivery Reliability

How agent replies are chunked and measured so Telegram, Slack, and Discord accept them intact

Every agent reply on Telegram, Slack, and Discord is split at paragraph boundaries, keeps its code fences intact, and is measured in the length unit each platform actually enforces.

from praisonaiagents import Agent
from praisonai.bots import Bot

agent = Agent(name="assistant", instructions="Answer with long, code-heavy replies.")
bot = Bot("telegram", agent=agent)
bot.run()

The user asks a question; the delivery layer chunks the reply so a long, code-heavy, or emoji-heavy answer reaches them whole rather than being rejected by the platform’s length limit.

Quick Start

1. Nothing to configure

   Chunking runs automatically on every adapter. A reply that fits the platform cap is sent as-is; a longer one is split at paragraph boundaries before sending. This applies to streamed replies too: the placeholder is edited with chunk 1 and the rest arrive as follow-on messages — streaming: true is all you need (PR #6081).

2. Code fences stay intact

   A fenced code block is kept whole rather than split mid-block, so users always see complete, copyable code.

3. Telegram measures in UTF-16

   Telegram counts its 4096-character cap in UTF-16 code units. PraisonAI counts in the same unit, so emoji- and CJK-heavy replies are chunked at the point Telegram actually enforces.

How It Works

chunk_message splits the reply, using the platform’s length_unit to measure each piece.

chunk_message(text, max_length, preserve_fences, length_unit) returns a list of chunks, each within the platform cap:

from praisonai.bots._chunk import chunk_message

# Telegram: 4096 cap, measured in UTF-16 code units
chunks = chunk_message(
    long_text,
    max_length=4096,
    preserve_fences=True,
    length_unit="utf16",
)
for chunk in chunks:
    await adapter.send_message(channel_id, chunk)

The adapter reads length_unit from its own platform_capabilities(), so you rarely call chunk_message yourself — it runs inside _send_long_message on every non-streamed reply, and inside DraftStreamer.finalize on every streamed reply.

Platform Caps

Common Patterns

Best Practices

Related