Plug any AI → Boot a Business Strategist, a Project Manager, a Data Analyst, a Research Director — from one portable workspace.
A persistent, structured operating layer that gives AGI-level Autonomy to any agent — Claude Code, Gemini CLI, Cursor, Codex, any harness.
For every AI power user who's done starting from scratch.
Did you PlugBoot your AI yet? 🔌
📝 Changelog: see CHANGELOG.md for the full, newest-first record of changes.
You have powerful AI. You also have a dozen projects, competing priorities, research piling up, and no way to make the AI remember what it worked on last Tuesday.
Every session starts from scratch. Every plan lives in a chat window. Every decision evaporates.
PlugBoot is a portable workspace you clone once and point any AI at. The AI lands in it and becomes a persistent, structured project manager that works for you — not for itself.
- It remembers everything. Plans, decisions, missions, research, toolboxes, and events all live in YAML files on your disk — not in the AI's context window. Resume any session in seconds.
- It works with any AI. Claude Code, Gemini CLI, Codex, Cursor, Hermes, Antigravity, any harness. The AI is a layer below the workspace. PlugBoot borrows the AI's reasoning; the workspace owns the state.
- It scales across projects. One workspace, unlimited projects. Each project gets its own board, missions, toolboxes, inbox, and data folder. Switch between them from a live dashboard.
- It runs in your browser. A single Python process serves both a sync engine and a real-time dashboard. No cloud, no account, no subscription.
A "project" can be anything: a business, a YouTube channel, a codebase, a legal-document workflow, a multi-account content operation.
| You are... | PlugBoot gives you... |
|---|---|
| A business owner managing growth, ops, and content | A structured project manager that never forgets |
| A data analyst juggling research and client deliverables | An inbox system that organizes external data into pillars |
| A project manager coordinating multiple streams | A living mission board with planning + execution lanes |
| An AI power user tired of starting fresh every session | Persistent memory that survives context windows |
PlugBoot/
AGENTS.md Agent boot authority — every harness reads this first
config.yaml Global control: which entities are active, automation levels
index.yaml Workspace map — every path, every entity, one file
_os/ THE ORCHESTRATOR (always on)
os-board.md Your OS identity and notes
os-runtime.yaml Live pillars, queues, objectives
os-missions.yaml Standard / research / evolution missions
os-toolboxes.yaml Toolbox registry
os-inbox.yaml Inbox + gateway tracker
os_prompts/ 10 hard laws the AI operates by
os-inbox/ Raw data drops + .<entity>-inbox_gateway/
your-project/ A PROJECT (repeat for each one)
*-board.md Project identity
*-runtime.yaml Live pillars and queues
*-missions.yaml All missions for this project
*-toolboxes.yaml Toolboxes for this project
*-inbox.yaml Inbox tracker
*-data/ Anything: code, docs, research, spreadsheets
.infra/
backend/ Sync daemon + dashboard server (Python, Starlette)
frontend/ Dashboard UI (htmx + Alpine + Cytoscape — no build step)
schemas/ YAML contracts (the law)
templates/ Board + mission templates
Three kinds of structured work:
- Standard — goals + ordered tasks. Supports rounds (repeating/persistent) for recurring workflows.
- Research — parameterized investigation. Set depth/detail/precision levels and sources (training data, web, YouTube, NotebookLM). Outputs topic trees with keywords and instructions.
- Evolution — the AI improves the workspace itself. Four modes: FAST (realtime intent), DEEP (full entity analytics), RESEARCH (from prior research), INBOX (from your data drops). Every run is gated by a readiness check so nothing advances until you approve.
Drop any file into a project's inbox folder — competitor research, reference docs, source data, anything. PlugBoot organizes it into a gateway under your project's pillars so the AI can find and act on it without re-reading everything every time.
Register agents and skills in a domain → toolbox → agent/skill hierarchy. Control what's active from the dashboard. The AI only uses what you've turned on.
# 1. Clone
git clone https://github.com/Auto-Skiller/plugboot.git my-workspace
cd my-workspace
# 2. Install dependencies
pip install -r .infra/backend/requirements.txt
# 3. Start the dashboard
py -3 .infra/backend/daemon.py
# → Dashboard at http://localhost:8000
# NOTE (Windows): use `py -3`, NOT `python`/`python3` — those resolve to the
# Windows Store alias and fail. See Hard Law 11.
# 4. PlugBoot your AI
# Open the workspace folder in Cursor, Claude Code, Gemini CLI, etc.
# The AI reads AGENTS.md and boots automatically.
# That's it. You just PlugBooted your AI. 🔌- Workspace owns state. Everything lives in YAMLs on your disk. The AI is a visitor, not the owner.
- Brain-first reading. YAMLs pre-describe every file so the AI picks up context without re-reading everything.
- No locks, no complexity. Simple writes, git is recovery.
- Content-aware sync. The daemon only writes files when real content changes — zero disk churn when nothing moves.
- One process. The sync daemon and dashboard server are a single Starlette process. No microservices, no orchestration layer.
- Convention now, MCP later. The harness bridge is convention-based today. A future MCP layer will expose the same read/write points as tools — without changing the model.
Two concepts that steer all AI work:
- Pillars are yours. Defined per project in its runtime YAML. They describe what matters to that project (e.g. "Audience Growth", "Revenue", "Operations").
- Aspects are fixed: Architecture, Capabilities, Monetization. They steer evolution and research runs so the AI focuses on the right dimension of improvement.
- MCP adapter layer (expose workspace as tools to any harness)
- Multi-user mode (shared workspace, per-user audit trail)
- NotebookLM gateway integration
- Hosted dashboard option (for teams without local Python)
- Project templates (e-commerce, content ops, SaaS, legal)
This project is in active development. Issues and PRs welcome. If you PlugBoot something interesting, open a discussion — we'd love to feature it.
🔌 PlugBoot — Plug any AI. Boot a project manager.