LLMunix Marketplace ALPHA
The LLMunix kernel as a Claude Code CLI plugin. Install the Pure Markdown Operating System into Claude Code via the marketplace and solve complex goals with the /llmunix command—dynamically creating specialized agent teams on the fly.
⚠️ Experimental Research - This project is a research prototype and will remain permanently in alpha status.
A Self-Evolving Kernel for Complex Problem Solving
LLMunix represents a paradigm shift in AI systems. Rather than shipping pre-built agents, it provides a kernel that creates agents on demand, learns from every execution, and evolves continuously by reusing successful strategies.
Core Components
🤖 Core System Agents
SystemAgent: High-level orchestrator for agent lifecycle, task decomposition, and workflow coordination
MemoryAnalysisAgent: Captures and logs system activities in structured memory traces
MemoryConsolidationAgent: Extracts learnings from logs and generates reusable patterns
đź“š System Specifications
SmartMemory.md: Hierarchical memory architecture (short-term logs, long-term learnings)
QueryMemoryTool.md: Search and retrieve knowledge from past projects
ClaudeCodeToolMap.md: Maps LLMunix operations to Claude Code's actual tools
⚡ Master Command
/llmunix: The kernel shell that analyzes goals, creates specialized agents, orchestrates execution, and consolidates learnings—all through a single command.
How It Works: Dynamic Evolution
1. Analyze the Goal
Decomposes your high-level objective into distinct tasks requiring specialized expertise (e.g., 'high-level vision', 'mathematical theory', 'quantum coding', 'technical documentation').
2. Create Project Structure
Generates a dedicated project workspace with components/agents/, output/, and memory/ directories for organized execution and learning.
3. Create Specialized Agents
This is the core innovation. For each required expertise, LLMunix writes a new agent markdown file with YAML frontmatter and detailed system prompts tailored to the specific project domain.
4. Orchestrate Execution
Delegates tasks to agents by reading their markdown definitions, invoking them via the Task tool, logging complete interactions, and managing dependencies between agents.
5. Produce Outputs
All deliverables (code, documentation, data files, visualizations) are saved to the project's output/ directory.
6. Learn and Improve
MemoryConsolidationAgent analyzes all short-term logs, extracts successful patterns, agent designs, and strategies, then generates project_learnings.md in long-term memory.
7. Continuous Evolution
Future projects query long-term memory to reuse proven agent designs, apply successful workflow patterns, avoid documented pitfalls, and bootstrap faster with adapted templates.
Installation
/plugin marketplace add evolving-agents-labs/llmunix-marketplace
/plugin install llmunix-plugin
Verify Installation
/llmunix --version
Start Using LLMunix
/llmunix "Create a biomedical quantum computing research project analyzing arterial flow patterns"
# LLMunix will:
# 1. Create projects/Project_aorta/
# 2. Generate VisionaryAgent, MathematicianAgent, QuantumEngineerAgent
# 3. Orchestrate their collaboration
# 4. Produce outputs and consolidate learnings
đź’ˇ Why This Matters
Traditional AI Systems: Ship hundreds of pre-built agents for specific domains. Limited coverage, generic solutions, systems grow bloated, no learning loop.
LLMunix Approach: Creates the exact agents you need with prompts tailored to your specific problem. Infinite domain coverage, problem-specific solutions, minimal core, continuous evolution.
🚀 Example: Dynamically Created Agents
For a biomedical quantum computing project, LLMunix might create:
- VisionaryAgent.md - Strategic conceptualization and high-level architecture
- MathematicianAgent.md - Theoretical framework and mathematical modeling
- QuantumEngineerAgent.md - Quantum circuit implementation and optimization
- TechnicalWriterAgent.md - Documentation and research paper generation
- CodeReviewerAgent.md - Quality assurance and code validation