Why this matters Open AI workflows fragment quickly: skills, connectors, and agent setups live in different tools and are hard to reuse across teammates or agent clients. This project treats those capabilities as first-class artifacts you can publish, assign, and invoke from any compatible MCP client while still offering a local desktop workspace for hands-on work.
What Sets It Apart
- Centralized capability distribution: publish skills, plugins, and model connections once in a control plane and assign them to organizations, teams, or individuals so the same capabilities appear in any MCP-capable agent.
- Agent-agnostic MCP integration: exposes a remote MCP endpoint so Codex, Claude Code, Cursor, OpenCode and other MCP clients can discover and execute the same skills without per-client rewiring.
- Local-first desktop + hosted workspaces: a full desktop app for macOS/Windows/Linux that can run work locally or hand it off to persistent hosted workspaces for reproducible environments and background tasks.
- Organization controls and security: admin-facing policies to manage desktop behavior, model provisioning, and per-user or shared connections, suitable for teams that must govern access.
Who it's for and tradeoffs
Great fit if you need to standardize agent capabilities across tools and teams, want a desktop environment that follows the same published skills, or need org-level controls for model access. Look elsewhere if you only need a single-agent integration, prefer a purely web-hosted SaaS without a local client, or require deeply sandboxed hosted execution today (hosted workspaces and long-running tasks are still maturing).
