Descriptions:
Stephanie Nyarko presents a step-by-step tutorial for building a nine-agent AI content production system using OpenClaw, self-hosted on AWS and operated entirely through Telegram. The architecture spans two coordinated teams: an ideation team led by an agent named Iris (supported by Spark for idea capture, Prism for scoring and prioritization, and Sketch for developing raw ideas into full briefs) and a production team led by Chief (with dedicated agents for research, first-draft writing in the creator’s voice, editing, multi-platform formatting, and publishing schedule generation).
The full workflow is triggered by a single Telegram message — a thought, a link, or a screenshot — which Iris’s team captures, ranks, and develops into a structured content brief, then hands off to Chief’s team for drafting and formatting across LinkedIn, Twitter, video scripts, email newsletters, and custom checklists. Nyarko walks through every configuration step: creating Telegram bots via BotFather, storing access tokens securely, editing OpenClaw’s YAML configuration file to support multiple simultaneous bot tokens, creating the agent directory structure on the AWS server, and writing Soul.MD files that define each agent’s personality, behavioral rules, and operational boundaries.
The Soul.MD configuration is highlighted as the mechanism that makes agents feel consistent and purposeful rather than generic — each agent has a defined character that persists across interactions. Nyarko makes her exact Soul.MD files available to members of her community. The tutorial assumes a pre-existing OpenClaw instance on AWS and cross-references her earlier infrastructure setup video for the initial provisioning steps.
📺 Source: Stephanie Nyarko · Published February 16, 2026
🏷️ Format: Hands On Build
