MadCop

A local-first AI agent desktop workstation.

MadCop is a cross-platform desktop application that brings the power of modern LLMs into a private, agentic workflow. It runs as a single Electron binary on macOS, Windows, and Linux, talks to any OpenAI-compatible API endpoint, and keeps your conversations, files, and knowledge base entirely on your machine. No cloud lock-in, no per-seat fees, no data leaving the device.

About

MadCop is an Electron + Vue 3 + FastAPI desktop agent that ships four integrated workflow systems end-to-end:

• Source-first creation — a planner → fetcher → outliner → writer pipeline that produces cited long-form articles backed by live web search.
• Proactive observer — watches file changes and terminal scrollback in the background, routes observations through a small LLM judge, and surfaces only what needs your attention.
• Knowledge canvas — an interactive graph (cytoscape + force layout) of the nodes your agent has learned about, with drag-to-pin positions and double-click-to-create.
• Auto-skill distillation — every long exchange (> 400 chars) is automatically captured as a reusable ~/.madcop/skills/<topic>.md so the agent gets smarter across sessions.

The single Electron binary contains both the Vue 3 renderer and the FastAPI backend (the backend is in-process for production and dev-mode-fast-iteration when the desktop loads it from a sidecar). Backend state lives in ~/.madcop/ (SQLite for memory, settings.json for providers, plan-mode-relevant flat files). The v4 agent engine emits a single normalized AgentStep stream so the SSE adapter is one switch statement, not five.

If you want to understand the shape of the system — the engine architecture, the memory model, the ProactiveObserver pipeline, the KnowledgeCanvas state machine — read ARCHITECTURE.md next.

This document explains the why behind the major design decisions — written for product managers and reviewers who want to understand how the system is put together, not just a list of features.

---

Skip to content

Preview

Clarify & choose Interactive question chips · User avatar · Task monitor panel

Structured report output Markdown tables · Key takeaways · Assistant avatar

Plan-driven execution Step-by-step planning progress · Live task tracking

---

What problem is MadCop solving?

The dominant LLM desktop clients (ChatGPT, Claude.ai, Gemini) are excellent chat surfaces but they assume a specific shape of interaction: one human, one model, one conversation at a time, with vendor-managed tools and memory. That works for "answer this question" but it does not work for "I need to (a) search the web, (b) read a local file, (c) summarise the result, (d) save a Markdown report to disk" — which is a normal afternoon for a product manager, analyst, or engineer.

MadCop is built around three observations: