LLMunix Starter ALPHA
A fresh LLMunix template optimized for Claude Code on the web. The factory for building specialized agents dynamically—no pre-built solutions, just the essential kernel to create exactly what you need, when you need it.
⚠️ Experimental Research - This project is a research prototype and will remain permanently in alpha status.
Philosophy: The Factory, Not the Products
Traditional AI systems pre-define agents for specific domains. LLMunix inverts this model—it creates agents on demand, tailored to your exact problem, learning from every execution.
❌ Traditional Approach
- ❌ Domain coverage is bounded
- ❌ Agents are generic
- ❌ Systems grow bloated
- ❌ No learning feedback loop
✅ LLMunix Approach
- ✅ Infinite domain coverage
- ✅ Problem-specific agents
- ✅ Minimal core (3 agents)
- ✅ Continuous evolution
What You Get
🎯 Essential Kernel Only
Only 3 core system agents: SystemAgent (orchestration), MemoryAnalysisAgent (pattern recognition), and MemoryConsolidationAgent (learning). No bloat, no pre-built domain agents.
🏭 Dynamic Agent Factory
Creates exactly the agents needed for your specific problem. Whether it's VisionaryAgent, MathematicianAgent, or QuantumEngineerAgent—generated on demand, tailored to requirements.
🧠 Continuous Learning
Every execution logs interactions to memory/short_term/, consolidates learnings to memory/long_term/, and uses past patterns to bootstrap future projects faster.
🌐 Web-Optimized
Designed specifically for Claude Code on the web. Connect your GitHub repository, give Claude a goal, and review PRs—all from your browser.
📝 Pure Markdown
Everything is markdown specifications. Agents, tools, memory—all defined in human-readable markdown files that Claude interprets at runtime.
🔐 Public or Private
Works with both public and private repositories. Configure GitHub App permissions, manage secrets securely, work on proprietary code in complete isolation.
Quick Start
Click "Use this template" on GitHub and create your repository (public or private)
2. Connect to Claude Code
Visit claude.ai/code and authorize GitHub access
3. Select your repository
Choose your new repository from the list
4. Give Claude a goal
Start with any ambitious, multi-faceted problem
5. Review and merge
Claude pushes changes to a branch for PR review
How It Works: Dynamic Evolution
↓
1. SystemAgent analyzes and decomposes the goal
↓
2. Creates project structure in projects/[ProjectName]/
↓
3. WRITES new agent markdown files for required expertise
↓
4. Reads each agent and invokes via Task tool
↓
5. Logs all interactions to memory/short_term/
↓
6. Produces final outputs in output/
↓
7. Consolidates learnings to memory/long_term/
↓
8. Future projects query and reuse these learnings
💡 Key Innovation
Instead of shipping hundreds of pre-built agents, LLMunix creates the exact agent you need with a system prompt tailored to your specific problem.
A biomedical quantum computing project? You get VisionaryAgent, MathematicianAgent, QuantumEngineerAgent. A web scraper? You get DataExtractorAgent, ParserAgent, ValidationAgent. Exactly what you need, nothing more.