This page documents the first Agentic Coding experiment. The goal is to shift away from coding and chatting inside a browser (e.g. Firefox), and instead work directly within a self-hosted environment powered by terminal agents, LiveCode UI, and Federated Wiki.
The pilot explores collaborative development, logging, and journaling through a blend of: - **LiveCode interface**: custom front-end - **Open Interpreter**: chat + code execution in terminal - **Xterm.js**: browser-based terminal widget - **CrewAI**: orchestration of local agents - **N8n**: automation and webhooks - **Federated Wiki**: writing, journaling, and embedding local project context - **Local assets folder**: for logs, chat transcripts, code snippets
All systems are hosted locally (no cloud dependencies), giving full control over the bootstrapping and memory process.
# Stack - Localhost:3000` → Federated Wiki server - Open Interpreter → accessible via terminal (`tty` or browser Xterm). See Xterm Conversations. - Live Asset Sandbox - CrewAI + N8n → background coordination, triggers, workflows
# Purpose
- Move development workflows **out of Firefox** into a **local terminal-first experience**
- Begin journaling and documenting coding sessions in **Federated Wiki**
- Start to form a **shared memory space** between agents and the user via persistent files
- Use **LiveCode UI** to experiment with launching commands, searching logs, and invoking workflows
- Lay groundwork for **agentic recall** — e.g., feeding past chat logs back into agents for context
## First Steps
### 1. Set up logging
- Create an `assets/chat/` folder inside your wiki directory
- Configure Open Interpreter (or a wrapper shell script) to append each input/output message to a Markdown file (e.g. `assets/chat/session-01.md`)
- Optional: Log timestamps for session replays or embedding in Wiki pages
### 2. Connect to Wiki
- Use LiveCode or a shell script to push chat excerpts or code snippets to your running Federated Wiki instance
- Start with simple Markdown appending to `assets/pages/*.json` or using the wiki’s HTTP API (if available)
- Alternatively, drag markdown logs manually into wiki pages for now
### 3. Launch Xterm with Interpreter
- Set up Xterm.js (via iframe or webview in LiveCode) to launch a session of Open Interpreter in a sandboxed shell
- Ensure terminal supports local path awareness so saved code can go to `assets/snippets/`
### 4. Start a chat session
- Begin chatting and coding with Open Interpreter
- Save every 10–20 lines of chat as a “scene” — use timestamps or headers to break it up
- Experiment with tagging: e.g. `## bugfix`, `## todo`, `## learn`, etc.
### 5. Create a live wiki page
- Create a page titled `Vibe Coding Pilot 01` and link it to:
- `[assets/chat/session-01.md chat log]`
- `[assets/snippets/init.js first snippet]`
- Optionally include:
- A screenshot of the terminal (or a TTY GIF)
- A LiveCode snapshot of the interface at launch
### 6. Begin building agent memory
- Start annotating messages or summaries in Wiki:
- “This solved the permissions bug”
- “Interpreter got stuck here”
- “Idea for future agent: auto-publish snippet on success”
## Possible Automations (via N8n or CrewAI)
- Auto-create a wiki page from each new chat log file
- CrewAI flow that reviews last 3 snippets and suggests refactor
- N8n webhook that listens for file creation in `assets/chat/` and appends a link to your pilot page
## Future Experiments
- Add YAML-like metadata to each chat log to track who/what/when
- Feed previous logs into Interpreter or Claude for smarter context
- Use shell agents to refactor or run snippets saved in `assets/snippets/`
- Try sending `LiveCode` scripts into the Interpreter for shell + GUI hybrid work
- github.com 
- github.com
- github.com
- xtermjs.org