Local-first desktop agent that herds Claude, Codex, and Goose

What it does

Wayland is a local-first desktop application built around a 47 MB Rust engine that acts as a command center for other AI agents like Claude Code, Codex, and Goose. It reads and writes your local files, executes shell commands inside a native per-OS sandbox, and maintains five SQLite-backed memory partitions that persist across sessions. Instead of resetting between chats, it keeps project state alive so you can cowork with it on multi-week tasks.

The interesting bit

When a task grows beyond one assistant, Wayland spins up a self-assembling team: a manager routes work to domain specialists that collaborate against a shared blackboard until the job ships. The engine also rewrites and re-scores its own skill prompts against an internal eval harness, so the system is meant to sharpen itself the more you use it.

Key highlights

• Drives a dozen AI CLIs from one interface, letting you switch models mid-task without juggling keys or contexts.
• Runs fully offline via Ollama or reaches for frontier cloud models; your files and shell never leave the machine.
• Ships with 88 pre-trained assistants, 1,974 skills, and 178 workflows, plus cron-style scheduling for recurring jobs.
• Sandboxed execution uses Landlock on Linux, sandbox-exec on macOS, and AppContainer on Windows.
• Licensed under AGPL-3.0 with no subscription or required backend.

Caveats

• The README openly discloses that the optional Flux Router inference backend is a paid service built by the same company. It is off by default and takes a separate key, but the vendor overlap is worth noting if you are wary of ecosystem bundling.

Verdict

Worth a look if you are drowning in disconnected AI CLIs and want a single, local-first workspace that remembers your projects across sessions. Skip it if you are looking for a simple chat wrapper or a fully browser-based tool.