Holus Content Engine
Holus is a multi-agent content engine that automates the full content lifecycle - from deciding what to create, to producing it, to learning from results. It coordinates specialized agents (writers, researchers, evaluators) that communicate through MCP tool calls to independent production silos: Pilaster for images, Genpeli for video, and a social media platform for publishing.
The Problem
Solo founders building multiple AI products can't keep up with content marketing across several platforms. Each product has a different audience, different formats, and different performance patterns. The work is repetitive but requires judgment - you need strategy, not just scheduling.
The Solution
An autonomous agent loop: OBSERVE what performed well via analytics → REASON about strategy using Claude Opus → ACT by dispatching to specialized content agents → EVALUATE every piece with domain-expert judges before publishing. The system learns weekly and adjusts its own strategy.
The Outcome
Content creation goes from manual and sporadic to continuous and data-driven. Every piece is quality-scored before publishing. The system improves its own prompts and strategy based on what actually performs.
Key Features
- Specialized agents organized by role: managers, specialists, evaluators, ops
- ReAct agent loop: observe → reason → act → evaluate
- 7 domain-expert quality judges with category-specific rubrics
- MCP integration with Pilaster, Genpeli, and Social Media silos
- Self-improvement loop: trajectory logging, weekly learning, strategy adaptation
- Three-layer prompt resolution for A/B testing agent prompts
Technology Stack
Interested in this project?
I'd love to discuss the technical details, challenges overcome, or similar projects I could build for you.
View Live AppLet's discuss this project