OpenClaw Telegram Setup guide

OpenClaw 3.2 adds streaming Telegram replies via BotFather

What actually changed

OpenClaw now streams agent output back to a Telegram chat instead of holding the user in a single "typing..." bubble. The host shows it returning a "long haikou" and a "short story" line-by-line — the creator's verdict: it "works really well" for long generations but the promo reel looked smoother than the real Mac/terminal run, which "decided to produce it in chunks."

BotFather token in 4 steps

1. Message @BotFather, hit /newbot, name it (creator used "box mining AI bot"), grab the token. 2. SSH into the box — creator used Terminal on macOS and Termius on iPad. Don't ask the agent: "it works 98% of the time, 2% of the time it'll kill itself." 3. Run openclaw update first. Target is OpenClaw 3.2. 4. Run openclaw config → channels → configure → telegram → paste token → enable Allow DM pairing → openclaw gateway restart.

Pairing quirk

After restart, DM the bot /start. It returns a pairing code; authorize it locally with openclaw tui and paste the code. The bot may also demand a name before pairing — creator bypassed with "please pair it now," after which it replied done.

Streaming tuning

If replies still arrive in chunks, ask the agent to flip the Telegram streaming setting from partial to on. The creator's testing shows streaming "works better for long messages" than short ones.

Known failure modes

• Token reset mid-shoot caused a pre-paired state in the demo; assume fresh setups skip straight to the chat screen.
• ~/.openclaw/config may need to be wiped (delete config → confirm) before re-running the wizard.
• AI-driven wiring via natural language is flaky — stick to the commands above.