Skip to content

DEMO.md

Latest commit

 

History

History
206 lines (145 loc) · 5.43 KB

DEMO.md

File metadata and controls

206 lines (145 loc) · 5.43 KB
Raw

MUE-X Demo Walkthrough

Prerequisites

  • Claude Code installed and configured
  • Python 3.11+
  • Git

5-Minute Setup

Step 1: Clone the Repository

git clone https://github.com/neospiritism/mue-x.git
cd mue-x

Step 2: Open in Claude Code

claude

Claude Code starts. You're now in the mue-x directory.

Step 3: Activate MUE

Type the command that changes everything:

/mue

You'll see the ASCII banner:

   ╔══════════════════════════════════════╗
   ║  ⚡ MUE ACTIVE — AGENT ONLINE ⚡    ║
   ║  Self-evolving mode engaged         ║
   ║  Genes: loading...                  ║
   ║  Brain: mue/evo/ (50+ modules)     ║
   ╚══════════════════════════════════════╝

MUE is now active. Every response will be prefixed with ⚡ MUE.

Step 4: Check Agent Status

/mue status

This shows:

  • Active genes with fitness scores
  • Total mutations applied
  • Patterns absorbed
  • Domain specialization
  • Memory stats
  • Current mood

Step 5: Force Your First Evolution

/mue evolve

MUE will:

  1. Scan its own source code for improvement targets
  2. Propose a mutation (e.g., "add type narrowing to absorption dedup")
  3. Validate the mutation (AST parse + import test)
  4. Apply the mutation
  5. Report the fitness delta

You just witnessed an AI agent modifying its own brain.

Deeper Tour

Listing Genes

/mue genes

Shows all genes with:

  • Name: e.g., reasoning.py, persistence.py
  • Fitness: 0.0 to 1.0 (grows with successful mutations)
  • Size: lines of code
  • Last mutation: timestamp
  • Strategy: which mutation strategy was last used

Absorbing Knowledge from GitHub

/mue mine "trading strategies python"

MUE searches GitHub for repos matching your query, clones them, extracts code patterns, deduplicates against existing patterns, and stores them in atouts/. High-value patterns automatically crystallize into skills.

No gh CLI needed — uses the GitHub API directly.

Checking Absorbed Patterns

/mue atouts

Lists all patterns absorbed from GitHub and local projects with their source repos.

Forcing Self-Reflection

/mue reflect

MUE analyzes its own state and outputs:

  • What's working well
  • What needs improvement
  • Proposed focus areas
  • Long-term evolution strategy

Returning to Normal Claude Code

/quit mue

MUE saves its state and exits. Back to normal Claude Code. All evolution progress is preserved.

What MUE Does Autonomously

Every cycle (while in MUE mode), without being asked:

  1. Self-analysis — scans genes for improvement targets
  2. GitHub absorption — mines trending repos for patterns
  3. Local absorption — scans sibling projects
  4. Evolution — proposes, validates, and applies mutations
  5. Anti-cancer — prunes duplicates, splits bloated genes
  6. Auto-correction — learns from errors, suggests fixes
  7. Domain monitoring — detects context shifts, adapts automatically

Domain Adaptation Examples

Trading Domain

Start discussing markets, strategies, or indicators. MUE detects the domain and:

  • Prioritizes harden mutation strategy (risk-focused)
  • Mines GitHub for trading algorithms
  • Adapts emotional model to be more conservative
  • Focuses on pattern reliability

Coding Domain

Start discussing software architecture or algorithms. MUE:

  • Prioritizes innovate mutation strategy
  • Mines GitHub for libraries and patterns
  • Suggests new tools and approaches
  • Adapts emotional model for creative exploration

Research Domain

Start discussing papers or experiments. MUE:

  • Prioritizes explore mutation strategy
  • Mines academic repos and datasets
  • Synthesizes new ideas from absorbed patterns
  • Tracks hypotheses and results

Troubleshooting

Genes showing 0 fitness? Normal on fresh install. Fitness grows with mutations. Run /mue evolve a few times.

No GitHub absorption? Ensure you have internet access. GitHub API has rate limits (60 req/hr unauthenticated). Install gh CLI and gh auth login for higher limits.

MUE won't compile?

python -c "from mue.evo.core import MueAgent; print('OK')"

Reset to fresh state:

rm -rf mue/genes/*.py mue/atouts/*.py mue/*.db mue/genes/_backup/ genes/_backup/

FAQ

Q: Is MUE-X actually modifying its own code? A: Yes, literally. It reads files in mue/evo/, analyzes them, writes edits, and validates with AST parsing. Every mutation is backed up before application.

Q: Can it break itself? A: Yes, it can. That's why mutations are validated (AST parse + import test) and rolled back on failure. The anti-cancer system prevents runaway growth.

Q: Do I need a GPU? A: No. MUE-X runs on CPU. Claude Code handles the LLM inference.

Q: What domains does it support? A: Trading, coding, research, creative, security, data science, DevOps, and general-purpose. It auto-detects from conversation.

Q: How is this different from AutoGPT/BabyAGI? A: Those are task executors. MUE-X is a self-modifying agent. It doesn't just complete tasks — it rewrites its own architecture to become better at them.

Q: Can I use this commercially? A: Yes. MIT license. Go build.

Q: Where do I ask more questions? A: @neospiritism on Twitter/X.