OpenClaw setup: Telegram replies with ElevenLabs voice

The speaker revisits OpenClaw with a practical task: drafting replies to relatives' tech support requests and returning voice memos in his own voice. Before the setup, he discusses security criticism. He cites Peter Steinberger's account of more than 1,100 security advisories, with about 650 resolved or closed. Those figures describe the account presented in the video, rather than establish that the remaining risks are fixed.

The tutorial uses a self-hosted installation on a Hostinger VPS. Hostinger sponsors the video, and the speaker chooses its manual quick start option so he can access the server over SSH. He connects a Telegram bot by creating it through Bot Father and entering its token in Hostinger's configuration.

On the server, the speaker locates the bot's personality instructions in the writable soul.md file. He uses an existing ElevenLabs voice profile and adds its API key and voice ID to an environment file. FFmpeg converts the generated MP3 audio into an .ogg voice memo, while tools.md gives OpenClaw context for the workflow.

The demonstrated AI agent analyzes a forwarded request, drafts a reply and passes it through a Python script to produce the memo. The speaker then forwards that memo to the relative. This setup depends on a hosted server and an external voice service; the tutorial does not demonstrate offline operation.