AI Automation: Deploying a Multi-Functional Agent System (Local & Analytical_Skills)

Automation Stack & Architecture

• Agent Framework: Bionic (via LM Studio running MLX or llama.cpp locally OR cloud models like GLM / Kimi).
• Integration Layers: Local file storage (auto-save functionality), Voice input processing (real-time speech recognition), external code editors (Claude Code, Codex, Cursor).
• Target Outcome: Automated coding, document management (PDF/Presentations), application deconstruction (Reverse Skill), and architectural explanation (ELI5 skillset).

Agent Roles & Tools Assignment

• Generalist Assistant (Bionic)
  • Goal: Write code, edit documents, create presentations and PDFs with automatic saving.
  • Tools: Real-time voice input (local audio processing), local LLMs managed by MLX or llama.cpp, Cloud fallback via GLM/Kimi APIs.
• Reverse Engineer Agent (Reverse Skill)
  • Goal: Analyze any type of software including APKs, websites, or programs to understand internal functions for reproduction in custom code.
  • Tools: Claude Code, Codex, or Cursor toolsets acting as an analytical engine.
• Architectural Explainer (Explainability Agent)
  • Goal: Deconstruct project architecture and explain complex logic like a beginner using visual HTML schemes.
  • Skills assigned: ELI5 (Explain Like I'm 5 / Explainer skillset).

Step-by-Step Workflow Orchestration

1. Initialize agentic environment locally through LM Studio without additional configuration if hardware permits; otherwise, switchto cloud models such as GLM or Kimi with Zero Data Retention protocol.
2. Input raw data either via text prompt, real-time local voice input, or by uploading files such as certain types of apps/codebases (e.g., APK, website source codes).
3. If task is reverse engineering: Deploy the specialized toolset (`Claude_Code` | `Codex` | `Cursor`) to scan application functions and extract underlying mechanics for code reproduction.
4. If task involves documentation/creation: Execute commands for writing/editing documents where any changes are passed back into storage automatically so results can be refined iteratively with the model.
5. For architectural deep-dives: Use explanatory prompts requesting an

! DYOR (Do Your Own Research)