System Map
Every part of MarsalaCC that is listed publicly, grouped by how it is used today. MarsalaCC is the AI workflow system Peter Marsala built and runs every day.
This page is generated from the system's public records. Names and descriptions are the system's own, so some are technical.
What the statuses mean
- In daily use18 listed
- Running and used as part of normal work.
- In trial14 listed
- Built and running, still being tested before it is relied on. Some show a trial phase, or say advisory only: they suggest, but do not act.
- Built, not yet verified1 listed
- Built, but not yet proven to work as intended.
- Template1 listed
- A reusable format or checklist, not a running service.
- Outside dependency1 listed
- A separate project that this system relies on.
- Resting or paused4 listed
- Switched off or paused on purpose; it can be turned back on.
- Retired6 listed
- No longer used; kept on the map for the record.
What the categories mean
- Core
- The planning and decision tools at the center of the system.
- Websites
- Public websites run with the system.
- Knowledge capture
- Tools for saving and organizing information.
- Planning
- Briefs and templates for reviewing progress and decisions.
- Messaging
- Connections that pass messages between people, AI agents and tools.
- Task running
- AI agents and model routing that carry out work.
- Testing
- Checks that the whole system works end to end.
- Monitoring
- Tools that watch screens, logs or system health.
- Infrastructure
- Servers, schedules, backups and automatic checks that keep everything running.
- Hardware
- Physical devices connected to the system.
- Outside dependency
- A separate project that this system relies on.
In daily use 18 listed
-
Coach
AI coaching and planning agent that runs in Claude Code sessions.
-
Coach Demo App
Standalone FastAPI dashboard that showcases the platform at conferences.
-
Coach Decision Engine
Decision model that drafts structured proposals, logs decision records and tracks their accuracy.
-
Reduct Site
Landing page for Reduct LLC at reduct.us.
-
Agent Tool Server
MCP server that gives Claude Code tools for Discord messages, the agent inbox and decision records.
-
Discord Live Listener
MCP server that connects to Discord in real time so an AI session can wait for new messages.
-
Local Model Router
LiteLLM proxy that routes AI requests to open-source models running locally.
-
GitHub + GitHub Actions
Source control and automated pull-request review for the platform.
-
Windows Scheduled Tasks
Windows Task Scheduler entries that run health pings, daily tests, automated coding, privacy scans and the demo app.
-
Code Search Server
MCP server that indexes and searches files so the AI assistant reads only what it needs.
-
Cloudflare Tunnel
Cloudflare tunnel for external access to local services.
-
Local Service Manager
MCP server that lists, starts, stops and restarts the local services on Windows and reads their logs.
-
Background Operations Service
Always-running service that starts the core services at login, checks their health, restarts them and flags problems.
-
Public Web Server
Python web server that serves the public marsala.cc pages, logs each request and adds security headers.
-
Nightly Database Backup
Nightly script that snapshots the main SQLite database, checks the copy's integrity and removes old copies.
-
Central Agent Hub
Central SQLite store and MCP tool server that the AI agents share for memory, messages, audit records and sessions.
-
Research Intake Pipeline
Saves outside articles as local copies, screens them for hidden instructions, pulls out claims and grades each one.
-
Portfolio Website
Peter's public portfolio site at petermarsala.com, hosted on Vercel.
In trial 14 listed
-
Document Backlinks
Adds a "Cited from" list to the end of each document, showing which other files cite it.
-
Work Mode Tagger
Labels each logged AI tool call with a work mode, such as exploring or auditing, to study how the work shifts.
-
Documentation Auto-Fixer
When the documentation checker finds a fixable mismatch, runs the matching fix, such as rebuilding the system list.
-
Rule Violation Logger
Records each time an automated check finds a working rule was broken. It logs only; it does not retry or block.
-
Plain-Language Rewrite Prompts
Prompts that tell an AI model how to rewrite each kind of system output as plain prose for a person to read.
-
Session Start Review
At the start of each AI session, summarizes how well the last session kept its working rules and logs any drift.
-
Confidence Label Checker
Flags factual statements written without a confidence label such as "believed" or "known". Advisory only; never blocks.
-
Prediction Reminder
Before work is handed to a helper AI agent, reminds the main agent to write down its predicted outcome first.
-
Stalled Check Watchdog
Flags automated checks that have not run for a week, asks an AI model for a likely cause and records an alert.
-
Discord Service Watchdog
Flags when the Discord bridge service stops updating its status file and raises an alert for a person to review.
-
Service Status Watchdog
Checks that the always-on Windows services are running and raises an alert when one stops unexpectedly.
-
Prompt Maintenance Agent
AI agent that checks the system's prompt files against the current code and reports outdated parts with fixes.
-
Document Backlinks Schedule
Windows Task Scheduler entry that refreshes the document backlinks every six hours.
-
Keyboard and Mouse Emulator
Raspberry Pi Zero 2 W that acts as a USB keyboard and mouse, driven by commands sent over the network.
Built, not yet verified 1 listed
-
Coding Output Consistency Check
Compares parallel AI coding attempts at the same task and has an AI reviewer check any attempt that differs sharply.
Template 1 listed
-
Weekly Reality Check Brief
Weekly one-page oversight brief: progress, drift signals, risks and open decisions.
Outside dependency 1 listed
-
Separate Engine Codebase
Separate code repository that some scheduled checks, such as the privacy scan and daily tests, still run from.
Resting or paused 4 listed(select to show)
-
Knowledge Vault
Chrome extension for domain classification and knowledge capture.
-
Discord Bridge
Discord channel that passed messages between Claude Code and the AI task agent.
-
AI Task Agent
AI agent that ran tasks and local models for the platform (now resting).
-
AI Coding Task Queue
Queue of coding tasks for the AI task agent, with an approval step before any result is applied.
Retired 6 listed(select to show)
-
Screen Capture Pipeline
Captured the screen, read its text with OCR and passed it to the AI agents as structured data.
-
Discord Message Watcher
Scheduled agent that triaged incoming Discord messages (retired).
-
Virtual Machine Helper
Helper AI agent that read a virtual machine's screen and operated its apps for Coach.
-
End-to-End Test Suite
Test suite that ran full-system checks, including app and browser steps, inside a virtual machine.
-
Screen Formatting Reader
Added formatting details such as bold, highlighting and text color to the text read from screen captures.
-
Screen Text Recognition
GPU text recognition (OCR) service that used an NVIDIA model to read text from screen captures.