Skip to content

Stars OS

The open-source operator layer for AI-native work.

Persistent memory, session lifecycle, and modular adapters for Claude Code, Cursor, and other AI agents — all CLI-first, local-by-default.

License: AGPL v3 Status: Pre-alpha

Website · Docs · Pro Services


The problem

You have 15+ Claude Code chats open across multiple projects. You don't remember which chat covered what. You forget where you left off. You switch projects ten times a day and lose the meta-overview each time.

You end up browsing ~/.claude/projects/ manually or grep'ing JSONL files to find the right thread.

Stars OS gives you

  • starsos alone — opens an interactive TUI cockpit (lazygit-style). The primary Game-Loop. Chats listed with simple aliases: A1, A2, B1, ...
  • starsos status — same view as one-shot output for scripts and piping
  • starsos chat resume A2 — implicit liftoff (git status, last session log, open tasks, Infisical hint) then spawn claude --resume in current terminal. --new-tab for new iTerm2 tab (macOS).
  • starsos chat task A2 "..." — dispatch a background subagent task to chat A2. Status visible in cockpit as A2 ▶ running. The killer feature for parallel orchestration.
  • starsos chat done A2 --status FERTIG — explicit touchdown: writes session log, updates your external registries via post-touchdown.sh
  • starsos chat tag/note/send/link — metadata on top of chats, JSONL stays untouched. Inbox-pattern for queued prompts.
  • starsos chat watch [--notify] — live updates + optional macOS notifications
  • starsos mcp install/call — call MCP servers (Paperclip, Moco, ...) without leaving the terminal
  • Hook system — ~/.starsos/hooks/post-touchdown.sh etc. let you update PROJECTS.md, ACTIVITY logs, People-CRMs, anything you own

Use . as shortcut for "current chat" when running inside Claude Code: starsos chat tag . wip.

All open source. All running locally on your machine. No account required. Stars OS is a filesystem orchestrator — it never makes LLM API calls of its own.


Install

You need Bun installed first:

curl -fsSL https://bun.sh/install | bash

Then install Stars OS:

bun install -g @starsos/cli

Or run without installing globally:

bunx @starsos/cli --help

Verify:

starsos --version

Homebrew formula and a single-binary installer (curl … | sh) are planned for v0.2.


Quick start

# 1. Initialize Stars OS
starsos init

# 2. Open the cockpit (interactive TUI — your home screen)
starsos

# In the cockpit you see your chats numbered: A1, A2, B1, ... grouped by project.
# Type:
#   A2 resume                 → liftoff + spawn `claude --resume` in this terminal
#   A2 new window             → open new iTerm2 tab with claude --resume
#   A2 task "fix the tests"   → dispatch background subagent (M1.5)
#   A2 tag wip                → add tag without switching
#   A2 done                   → touchdown with summary prompt
#   r                          → refresh
#   q                          → quit
#
# Or one-shot from the shell:
starsos chat resume A2 --new-tab
starsos chat tag A2 wip
starsos chat send A2 --prompt "remember to update the README"

That's the Game-Loop. Projects are auto-detected from your chats' working directories. Liftoff and touchdown wrap each chat session. Your own ~/.starsos/hooks/post-touchdown.sh keeps your external registries (PROJECTS.md, ACTIVITY-LOG.md, etc.) in sync. Everything else (MCP servers, migrations, task subagents) builds on top.


How it works

┌─────────────────────────────────────────────┐
│  Your terminal (Claude Code, Cursor, …)     │
└──────────────────┬──────────────────────────┘
                   │
                   ▼
┌─────────────────────────────────────────────┐
│  starsos CLI                                │
│  ┌──────────┐ ┌─────────┐ ┌─────────────┐  │
│  │ Sessions │ │ Memory  │ │ Projects    │  │
│  └──────────┘ └─────────┘ └─────────────┘  │
│  ┌──────────────────────────────────────┐  │
│  │ Adapter SDK (plug in anything)       │  │
│  └──────────────────────────────────────┘  │
└──────────────────┬──────────────────────────┘
                   │
                   ▼
            ~/.starsos/  ← All your data, on your disk

Local-first. Stars OS metadata lives in ~/.starsos/. Chats stay where Claude Code put them (~/.claude/projects/*/*.jsonl) — Stars OS reads but never writes them.

MCP-native. Adapters are standard MCP servers (@starsos/mcp-server-*). The same servers you install for Stars OS also work in Claude Desktop, Cursor, Continue.dev.

Filesystem-orchestrator only. Stars OS does not make LLM API calls itself. It coordinates Claude Code, not duplicates it.


Adapters

All adapters are standard MCP servers. They work in Stars OS, Claude Desktop, Cursor, Continue.dev — anywhere that speaks MCP.

MCP server Purpose Milestone
@starsos/mcp-server-paperclip Project list, activities, time logging M4 wedge
@starsos/mcp-server-moco Time tracking sync M5
@starsos/mcp-server-clickup Tasks and time entries M6
@starsos/mcp-server-hetzner Cloud host management Beyond M9
@starsos/mcp-server-cloudways App management Beyond M9

Write your own MCP server using the official @modelcontextprotocol/sdk. Stars OS does not require an SDK of its own.


Why open source

We believe the code layer is a commodity. The lasting value is in:

  • People networks — who you know, who you trust
  • Systems — workflows you've earned through experience
  • Data — datasets you've gathered or refined
  • Real-world infrastructure access — credentialed, audited, accountable

Stars OS is AGPL-3 forever for the core. We never ask you for an account to use it. If you want a managed hosted version with team views, audit logs, and managed sync, see starsos.dev.


Roadmap

Milestones, not dates. We ship when a milestone is real.

  • M1 — Chat Cockpit + Liftoff/Touchdown + Hooks + TUI: interactive cockpit, letter+number aliases, implicit liftoff at resume, explicit touchdown at done, hook system, --new-tab
  • M1.5 — Task Subagent: chat task <alias> "..." dispatches background subagents, status visible in cockpit
  • M2 — Project Registry (minimal): pretty-name layer over auto-detected projects
  • M3 — StarsHub Migration: import project names and external refs (sessions deliberately skipped — chats are sessions)
  • M4 — MCP Client + Wedge Adapter (Paperclip, contingent on maintainer verification)
  • M5 — Moco adapter as MCP server
  • M6 — ClickUp adapter as MCP server
  • M7 — Public push (silent mode, no Show HN)
  • M8 — Dogfood-heavy phase
  • M9 — External wedge validation (3–5 operators)
  • Beyond M9 — Show HN (conditional), Liftoff/Touchdown, memory subsystem, more adapters

See STABILITY.md for what's guaranteed vs. what may move. docs/adr/ holds the architectural decisions.


Community

A Discord server will open closer to v0.1 launch.


Contributing

We welcome contributions, especially:

  • New adapters for tools and platforms you use daily
  • Documentation improvements
  • Bug reports with reproducible steps
  • Recipes and workflow examples

Read CONTRIBUTING.md before opening a PR.


License

AGPL-3.0 for the core. Pro services on starsos.dev are under separate commercial terms.

Inspired by Paperclip — both in spirit and in license choice.

Built by Stars Media IT GmbH and the community.

About

The open-source operator layer for AI-native operators. Persistent memory, session lifecycle, and modular adapters for Claude Code.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages