🎯 Background & Motivation

In the generative AI era, relying on manual prompting or single-agent monolithic architectures fails to sustain high-frequency, high-quality content pipelines across multiple channels.

Key limitations of single-agent approaches:

  1. Role Overload: Forcing a single prompt to handle topic discovery, drafting, fact-checking, and formatting results in generic, repetitive outputs.
  2. Lack of Feedback Loops: Without dedicated critique and refinement agents, content lacks depth and human nuance.
  3. Disconnected Dispatching: Generated outputs are isolated from scheduling queues and publishing pipelines.

OnlyBots was designed to explore autonomous multi-agent collaborative workflows: Simulating an editorial pipeline with specialized agents (Discovery, Creator, Critic, Formatter) to achieve end-to-end automated content synthesis.

🚀 System Architecture & Key Features

  • 🤖 Specialized Multi-Agent Role Orchestration:

    • Discovery Agent: Scans technical streams around the clock, clustering discussions and identifying high-engagement core topics.
    • Creator Agent: Leverages Qwen and Hermes LLMs to craft contextual, in-depth narratives tailored to target developer personas.
    • Critic Agent: Reviews draft outputs for factual consistency, logical flow, readability, and de-AI phrasing.

  • 📦 Intelligent Candidate Queue & Dispatcher:

    • Automatically files vetted content into an interactive staging queue, supporting 1-click manual reviews or autonomous scheduled publishing.

  • 🔌 Extensible Tool & Protocol Integration:

    • Integrates with MCP (Model Context Protocol) tool servers, enabling agents to dynamically invoke external search, data lookups, and specialized sub-tasks.

🏗️ Multi-Agent Pipeline Topology

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16

[ Global Technical Discussion & News Feeds ]
       │
       ▼
[ 1. Discovery Agent (Topic Clustering & Relevance Filtering) ]
       │ (Selected High-Value Core Topics)
       ▼
[ 2. Creator Agent (Contextual Drafting / Qwen & Hermes LLMs) ]
       │ (Initial Draft)
       ▼
[ 3. Critic & Fact-Check Agent (Verification & Nuanced Polishing) ]
       │ (Quality-Scored & Approved)
       ▼
[ 4. Dispatcher Pipeline (Candidate Queue & Staging Box) ]
       │
       ▼
[ Visual Management Dashboard & Multi-Channel Distribution ]

🛠️ Engineering Highlights

  1. State Machine & Decoupled Messaging: Each agent maintains isolated contextual memory and role boundaries, communicating asynchronously over an internal event bus.
  2. Heterogeneous Model Routing: Routes high-volume filtering to lightweight models while reserving reasoning-heavy models for creative drafting and critique, optimizing cost and latency.

📈 Implementation Status

  • Multi-agent orchestration engine & event messaging bus
  • Customized drafting and verification workflows with Qwen & Hermes
  • Interactive candidate staging queue and multi-platform scheduler
  • Stable 24/7 self-hosted HomeLab deployment

🔗 Project Status

  • Type: Proprietary multi-agent automation platform
  • Deployment: Running privately on self-hosted infrastructure