This project sets out to build a local, educational agentic development environment. It integrates the following components, all running on a personal laptop: - **Federated Wiki** (local instance): A creative knowledge-building and storytelling tool - **LiveCode app**: A custom UI and control surface for scripting, coordination, and educational interactions - **Agentic tools**: A set of modular agents (e.g. using CrewAI, Claude Code, Open Interpreter) - **Assets folder**: A shared directory used by Federated Wiki to host and display custom mini‑apps
# Workflow Description The system is structured around the following loop: 1. **Federated Wiki hosts context and goals** - Pages describe problems, tutorials, or prompts - Each page can embed output using the `frame` plugin to display HTML/JS artifacts 2. **Agents read/write to the assets folder** - Tools generate or edit HTML and JavaScript mini‑apps inside the `assets/` folder - These are instantly served by Federated Wiki and viewable via the browser 3. **LiveCode connects UI and logic** - Executes terminal commands - Displays agent status and lets the user interact with workflows - Provides control over `pyenv`, Claude, Open Interpreter, etc. 4. **User plays an active role** - Agents suggest code, write files, and execute tests - The user validates, reruns, or refines agent outputs - Results are committed back to the wiki as knowledge (or shared via Git/Federation) 5. **Result: Human + AI co-creation of educational content** - Mini-apps, guides, templates, and systems are all built within this hybrid loop - Wiki pages record both the knowledge and the tools to reproduce it
# Educational Goals - Create a **self-sufficient agentic lab** that runs entirely on local hardware - Help users set up their own agents and contribute to a decentralized network - Produce interactive “**Guides**” using LiveCode + Wiki that show others how to participate - Encourage experimentation with lightweight HTML/JS “vibe code” apps as learning projects - Preserve all components (data, code, history) in federated, remixable wiki pages
# ✅ Strengths - **Elegant use of Federated Wiki** as both UI and structured memory - **No need for external servers** — the assets directory and frame plugin create a seamless loop - **Agent productivity is tightly scoped** — building simple HTML/JS apps is an achievable goal for LLMs - **LiveCode provides direct interaction** — offering buttons, prompts, and feedback outside the browser - **Strong educational framing** — packaging results as “Guides” helps others follow and join in
# ⚠️ Challenges & Areas to Improve - **Lack of orchestration**: Agent workflows may become fragmented. Consider a lightweight orchestrator (e.g. LangGraph or n8n) to manage flows. - **Agent coordination**: Who plans vs who codes vs who validates? Define agent roles clearly (e.g. Planner, Coder, Validator). - **No stateful memory yet**: Agents won’t know what was done previously unless you implement state (via wiki page history or LiveCode session tracking). - **Security and sandboxing**: Allowing agents to write files and run scripts needs safeguards — especially if shared with others - **LiveCode ↔ Agent I/O bridge**: Consider how LiveCode can pass instructions, display agent output, or trigger subprocesses consistently
# Next Steps 1. Build or install the following base tools: - Open Interpreter or Claude Code for command line code reasoning - CrewAI or AutoGen for role-based agent delegation - LangGraph or n8n if orchestration is desired 2. Wrap agents to target the `assets/` folder: - Provide prompts like: "Build a simple HTML5 counter that loads in the frame plugin" - Write to `wiki/assets/my-app.html` and reference from a test wiki page 3. Extend LiveCode to: - Trigger workflows or model selection (e.g. Claude vs local model via Ollama) - Watch for file changes or completion signals from agents - Display logs, execution status, or feedback prompts 4. Create a wiki page template for a **Guide**: - Description, task, and embedded app in frame - Links to assets and explanation of agent steps - Yam metadata or federated journal to track changes and forks
This system provides a playful, remixable, and entirely local way to explore agentic computing. It’s ideal for education, storytelling, tool-building, and onboarding others into a federated AI world.
# See - Federated Plan - Agentic Routers and Federated Agents - Hitchhiker Agents and Agent Roles