Wayland: Local-First AI Agent Command Center

FerroxLabs/wayland provides a local-first desktop AI agent designed to unify and orchestrate multiple command-line interface (CLI) AI tools. It targets developers, researchers, and power users seeking a centralized control plane for diverse AI models (Claude Code, Codex, Gemini, Qwen, etc.) directly on their machine. The primary benefit is enhanced productivity and control by consolidating AI workflows, ensuring data privacy, and enabling complex, multi-agent projects without vendor lock-in or cloud dependencies.

How It Works

Wayland operates via a four-step cognitive loop: Perceive, Reason, Act, and Evolve. It functions as a local command center, running on the user's machine and leveraging native OS sandboxing for security. The system orchestrates various AI CLIs, maintaining shared memory and state across sessions and agents. Tasks are executed using built-in tools (file operations, shell commands) and external connectors, with an optional "Evolve" step that refines skill prompts. The core Wayland-Core engine is a single Rust binary, simplifying deployment.

Quick Start & Requirements

• Primary Install: Download native installers (.dmg, .exe, .deb) from the Releases page. For Debian/Ubuntu: sudo apt install ./Wayland-*.deb. For standalone CLI: npm install -g @ferroxlabs/wayland-core or npx @ferroxlabs/wayland-core.
• Prerequisites: Node.js (v18+ for self-host, v22-24 for build), Bun (v1.3+ for build).
• Notes: macOS and Windows may show security warnings on first launch due to lack of notarization/code-signing; manual overrides are required. Native installers bundle the Wayland-Core engine.
• Links: Website and download links are referenced but specific URLs are not provided in the README text.

Highlighted Details

• Unified Agent Control: Manages numerous AI CLIs (Claude Code, Codex, Gemini, Qwen, Goose, etc.) from a single interface.
• Local-First & Privacy: Executes on the user's machine, utilizing local keys, files, and shell within native sandboxes (Landlock, sandbox-exec, AppContainer). Supports fully offline operation with local models (e.g., Ollama).
• Persistent State: Maintains project state, files, and memory across sessions using SQLite-backed storage.
• Workflows & Teams: Includes 177 pre-built workflows and supports self-assembling teams of AI specialists for complex tasks.
• Integrations: Connects to external services (Gmail, Slack, Asana) and supports any Model Context Protocol (MCP) server.
• Built-in Image Generation: Integrated image generation capabilities.
• Scheduled Tasks: Enables scheduling of recurring jobs.
• Constitution Enforcement: Enforces user-defined rules via ~/.wayland/CONSTITUTION.md.

Maintenance & Community

No specific details regarding maintainers, community channels (Discord/Slack), or project health signals beyond the open-source nature were found in the provided text.

Licensing & Compatibility

• License: GNU AGPL-3.0 for both the application and the Wayland-Core engine.
• Compatibility: Allows self-hosting, modification, and commercial use, provided that any networked service derived from Wayland must also be released under the AGPL-3.0 license. The "Wayland" name is trademarked; forks must adopt a different name but can indicate compatibility.

Limitations & Caveats

The AGPL-3.0 license mandates that any networked service built upon Wayland must publish its source code under the same terms. Forked projects are required to use a distinct name due to trademark restrictions. Initial releases on macOS and Windows may trigger security warnings requiring manual user intervention.