内容

<a href="https://getwayland.com"> <img src=".github/assets/hero.png" alt="Wayland. One agent. Every agent. The local AI agent that drives every other AI on your machine." width="100%"/> </a> Plan, build, and ship anything from one place.  One agent. Every agent. Wayland is a local-first desktop AI agent that runs on your machine and drives Claude Code, Codex, Gemini, Qwen, Goose and a dozen more from one command center. Your keys, your files, your shell. Stop juggling five CLIs. <a href="https://github.com/ferroxlabs/wayland/releases"><img src="https://img.shields.io/github/v/release/ferroxlabs/wayland?style=flat&color=ff6b35&label=release" alt="Release"/></a> <a href="https://github.com/ferroxlabs/wayland/releases"><img src="https://img.shields.io/badge/platform-macOS%20%7C%20Windows%20%7C%20Linux-08070c?style=flat" alt="Platforms"/></a> <a href="./LICENSE"><img src="https://img.shields.io/badge/license-AGPL--3.0-08070c?style=flat" alt="License: AGPL-3.0"/></a> <a href="https://www.rust-lang.org"><img src="https://img.shields.io/badge/engine-Rust-08070c?style=flat" alt="Engine: Rust"/></a> <a href="https://github.com/ferroxlabs/wayland/releases/latest">Download</a>  ·  <a href="https://getwayland.com">Website</a>  ·  <a href="https://docs.getwayland.com">Docs</a>  ·  <a href="#build-from-source">Build from source</a>  ·  <a href="#supported-agents-and-models">Supported agents</a> ## What is Wayland Wayland is the command center for every AI CLI on your machine. It is a full agent in its own right, and it also drives the CLIs you already use, all from one app, all on your own keys. It perceives, reasons, acts, and evolves locally. Install it, paste one key, and get real work done while one agent plans, builds, and ships across every AI tool you already pay for. Drive Claude Code, Codex, Gemini, Qwen, Goose and more from a single command center. Switch models mid-task. Keep your files, your shell, and your keys local. You cowork with the system to get whole jobs done: write the book, run the multi-week project, take an idea all the way to shipped. You set the goal and stay in control while it handles the grunt work, calls in specialists, and keeps the thread across every session. Free and open source, app and engine both. No subscription, no cloud round-trip, no lock-in. ## Cowork: plan, build, and ship anything from one place AI finally gets useful when it stops answering one question at a time and starts doing the whole job. That is Cowork: you and the system work the problem together, from the first rough idea to the shipped result. Write the book. Run the multi-week project. Plan, build, and ship the thing you actually came to do. It is collaborative and autonomous at the same time. You set the goal and stay in control. Wayland keeps state across the whole arc of the work: your files, the running plan, and a memory that does not reset between sessions. When a job is bigger than one chat, it scales up on its own: - **One assistant, real file access.** Cowork operates directly on your files with the built-in tools (Read, Write, Edit, Bash, Grep, Glob), processes real documents, and plans multi-step work. Start in a read-only Plan mode to see the approach before anything is touched. - **A self-assembling team when the job is big.** Hand the goal to a team. A manager you talk to in plain language assembles the right specialists and they work the task together against one shared blackboard until it is done. Keep a Standing Company on call, or spin up an ad-hoc team for a single push. - **Pre-trained specialists on tap.** Pull in a domain expert for the part that needs one, then drop back to driving the whole thing yourself. - **It keeps going on a schedule.** Put the recurring parts on cron and they run while you are away. You are not prompting a chatbot. You are coworking with a system that remembers the project, owns the grunt work, and checks in when it matters. ## Why Wayland **One team, not isolated CLIs.** Your CLIs are brilliant strangers who never met. Each one forgets everything between sessions and lives in its own silo. Wayland is the command center that makes them one team: shared memory across every agent, one workflow from idea to ship, multi-AI cross-audit, and opt-in [Flux Router](https://fluxrouter.ai) routing that sends each task to the model that fits it. **Your machine, not someone's cloud.** Chatbots run on someone else's servers, behind someone else's keys and retention policy. Wayland runs on your machine, reads and writes your files, and runs your shell commands through the bundled engine, inside a native per-OS sandbox behind a single egress chokepoint (bubblewrap on Linux, sandbox-exec on macOS; on Windows a kill-on-close Job Object by default, with AppContainer available via `WAYLAND_SANDBOX=appcontainer`). Tools the desktop app dispatches itself run with your account's privileges — [the desktop security model](https://docs.getwayland.com/concepts/security-model-desktop/) says exactly which is which. Go fully local with Ollama, or reach for a frontier cloud model when you want it. **A system, not a prettier chat window.** Most are a nicer window onto one model, and they reset every time you close them. Wayland is a system that compounds: an engine that rewrites and re-scores its own skill prompts against an eval harness, a memory that persists across sessions (five SQLite-backed partitions), plus ready-made assistants, self-assembling teams, workflows, and schedules that run without you. It gets sharper the more you run it. ## Download Grab the latest build for your platform. No account, no sign-up. Every link opens the latest Releases page, where you pick the file for your platform. | Platform | Architecture | File | | ----------- | -------------------------------- | ------------------------------------------------------------------ | | **macOS** | Apple Silicon (M1 and up) | [.dmg](https://github.com/ferroxlabs/wayland/releases/latest) | | **macOS** | Intel | [.dmg](https://github.com/ferroxlabs/wayland/releases/latest) | | **Windows** | x64 | [.exe](https://github.com/ferroxlabs/wayland/releases/latest) | | **Windows** | ARM64 | [.exe](https://github.com/ferroxlabs/wayland/releases/latest) | | **Linux** | x64 (Debian / Ubuntu) | [.deb](https://github.com/ferroxlabs/wayland/releases/latest) | | **Linux** | ARM64 (Debian / Ubuntu) | [.deb](https://github.com/ferroxlabs/wayland/releases/latest) | | **Linux** | x64 (Fedora / RHEL / openSUSE) | [.rpm](https://github.com/ferroxlabs/wayland/releases/latest) | | **Linux** | ARM64 (Fedora / RHEL / openSUSE) | [.rpm](https://github.com/ferroxlabs/wayland/releases/latest) | | **Linux** | x64 (any distro) | [.AppImage](https://github.com/ferroxlabs/wayland/releases/latest) | | **Linux** | ARM64 (any distro) | [.AppImage](https://github.com/ferroxlabs/wayland/releases/latest) | The installer bundles the Fuigo engine for your platform, so a clean install runs agents the moment you connect a Flux Router key. ### First launch <details> <summary>macOS says Wayland cannot be opened?</summary> Current releases are signed and notarized, so they should open normally. If you still see this, you are almost certainly on an older download, or the file lost its quarantine attributes in transit. To open it, go to **System Settings, then Privacy and Security**, scroll down, and click **Open Anyway** next to the Wayland entry, then confirm with **Open**. On older macOS, right-click the app in Applications and choose **Open, then Open**. </details> <details> <summary>Windows SmartScreen warning?</summary> SmartScreen warns on any publisher it has not seen enough downloads from yet, signed or not. Click **More info, then Run anyway**. The build is the unmodified release artifact. </details> <details> <summary>Linux install</summary> ```bash # Debian / Ubuntu sudo apt install ./Wayland-*.deb ``` Linux builds target glibc, not musl. </details> Prefer to build it yourself? See [Build from source](#build-from-source). ## Self-host in the cloud Run Wayland as an always-on agent on any Linux box or VPS, reachable from your phone. Three commands, no config files. > **Live on npm.** `getwayland` is published (`latest` is 0.12.4). The flow below is the verified one: it boots on a fresh Ubuntu VPS and answers through Flux. > > **Disclosure:** Flux Router is also built by Ferrox Labs. It is a paid inference router, and we would rather say so here than have you find out later. It is optional, off by default, takes its own separate key, and Wayland has no required backend of any kind — any OpenAI, Anthropic, Gemini, or local Ollama key works exactly as well. Delete the Flux key and everything else keeps working. **Requires Node 18 or newer.** On a fresh VPS: `sudo apt-get update && sudo apt-get install -y nodejs npm` ```bash npm install -g getwayland wayland setup # paste a Flux key (free at fluxrouter.ai) or any OpenAI, Anthropic, or Gemini key wayland start # prints a QR code and admin login in your terminal ``` Scan the QR from your phone and log in. Reach it over your tailnet (below), not a public IP. No reverse proxy, no DNS, no dashboard to configure. - **Keep it private (recommended).** Put it behind [Tailscale](https://tailscale.com) so it never touches the public internet. `tailscale serve 3000` gives you HTTPS, tailnet-only. - **Run it 24/7.** `wayland setup` offers to install a systemd service that survives reboots. - **No key?** Grab a free [Flux Router](https://fluxrouter.ai) account: one key, every model, best-fit routing. Cloud self-host runs the headless server and routes through your provider key or Flux. The desktop app, with the bundled Fuigo engine, voice, and image generation, ships as the native installers above. ## Features ### One agent commands them all Drive Claude Code, Codex, Gemini, Qwen, Goose and a dozen more from a single command center, on your own keys, never locked into one provider. One login, one system, one place to work. <img src=".github/assets/dashboard.jpg" alt="Wayland desktop dashboard, the command center for every agent, with assistants, workflows, and teams" width="100%"/> ### Skip the cold start with ready-made specialists Instead of coaching a blank chatbot, launch a specialist that already knows the domain: research, writing, sales, operations, strategy. Each one carries a curated skill set, so it works like an expert from the first message. Launch one in a click, build your own, or filter the whole library by type and domain. <img src=".github/assets/assistants.jpg" alt="Wayland assistants library showing pre-trained domain specialists in a browse grid" width="100%"/> ### Workflows from idea to outcome 178 ready-to-run workflows that take you from a blank page to a finished deliverable: a launch plan, a competitor teardown, a month of content, a release write-up. Run one as-is, build your own, or put it on a schedule. Each one walks the steps so you get the outcome, not just a starting point. <img src=".github/assets/workflows.jpg" alt="Wayland workflow running step by step toward a finished outcome" width="100%"/> ### Self-assembling teams Ready-made teams of AI specialists, each led by a manager you talk to in plain language. They self-assemble around your goal and work autonomously against one shared blackboard until the job is done. <img src=".github/assets/teams.jpg" alt="Wayland multi-agent team coordinating on a shared blackboard" width="100%"/> <img src=".github/assets/teams-in-action.jpg" alt="A Wayland team in action: a leader coordinating X, LinkedIn, and Facebook specialists plus a Humanizer, each running its own model, working the task live across shared columns" width="100%"/> ### Projects that keep work moving Pick up a multi-week effort where you left it. A project keeps the chats, files, reference material, instructions, and scoped memory for one piece of work in a single place, so the context is still there when you come back days later. No re-briefing, no lost thread. Organized, shareable, and ready to ship. <img src=".github/assets/projects.jpg" alt="Wayland project workspace with scoped files, chats, and memory" width="100%"/> ### Schedule recurring work in plain language Wake up to finished work. Tell Wayland what to run and when (daily, weekly, month-end) and each run delivers its result to its own conversation, done and waiting. The standup digest, the weekly report, the month-end close: handled before you sit down. <img src=".github/assets/scheduled-tasks.jpg" alt="Wayland scheduled task configuration with recurring runs" width="100%"/> ### One live pane for everything running Stop wondering what your agents are doing. Mission Control puts every team task and scheduled job in one live command center, grouped by what needs you first: what's failing, what's active, what's queued. Real-time status, owners, and next-run times, so nothing runs silently and nothing slips. <img src=".github/assets/mission-control.jpg" alt="Wayland Mission Control, a live command center showing every team task and scheduled job grouped by status with real-time counts" width="100%"/> ### It evolves. It does not forget. Stop re-explaining yourself. Wayland remembers your stack, your decisions, and how you like to work, and it shares that memory across every agent and project. Teach it once and it stays taught. Under the hood: a cognitive memory with five partitions (working, episodic, semantic, procedural, and a model of you) across three tiers (session, project, global), SQLite-backed, not a scratch file. It gets sharper the more you run it. <img src=".github/assets/memory.jpg" alt="Wayland unified memory view shared across every agent and project" width="100%"/> ### Connect your entire stack Let your agent act where your work lives: send the Gmail, file the Stripe invoice, post to Slack, update the Asana task, across Google Workspace, Microsoft 365 and more. Add any Model Context Protocol server over stdio, SSE, or streamable-HTTP. Your stack becomes the agent's hands, not a copy-paste chore. <img src=".github/assets/mcp.jpg" alt="Wayland MCP integrations browser connecting external tools" width="100%"/> ### Generate images in the same place you work Image generation is a built-in tool, wired to multiple providers on your own keys. Make the launch graphic, the deck visual, or the mockup inline with the rest of the job, no separate app, no context switch. ### Run it from anywhere Your desktop keeps working after you walk away from it. Scan a QR code to check in and drive everything from your phone or any browser, or just text your agents and teams like you would text a coworker from Slack, Telegram, Discord, WhatsApp, Signal, SMS, iMessage, Teams, or email. 25 channels in all. Kick off a job from the train, read the result over coffee. <img src=".github/assets/remote-webui.jpg" alt="Wayland remote web UI controlled from a phone over the browser" width="100%"/> ### Govern every turn with a Constitution Set your rules once and every agent follows them, no matter which CLI runs the task. Coding standards, tone, things to never touch: write them in plain English in `~/.wayland/CONSTITUTION.md` and Wayland enforces them on every turn, with per-specialist overrides when one needs different rules. One source of truth your whole fleet obeys. <img src=".github/assets/constitution.jpg" alt="Wayland constitution rulebook editor that governs every turn" width="100%"/> ## Supported agents and models Wayland spawns each CLI in [ACP](https://agentclientprotocol.com) mode and you bring the CLI's own auth. The bundled Fuigo engine speaks ACP too, with no install and no login of its own; Gemini runs natively. | Agent | Command | Connect with | | :------------------------------------------------------------------------------------------------------- | :----------- | :---------------------------------- | | <img src=".github/assets/logos/openclaw.svg" width="20" valign="middle"/>  **OpenClaw** | gateway | OpenClaw account | | <img src=".github/assets/logos/hermes.svg" width="20" valign="middle"/>  **Hermes** (Nous Research) | `hermes acp` | Hermes login | | <img src=".github/assets/logos/claude.svg" width="20" valign="middle"/>  **Claude Code** | `claude` | Claude login or `ANTHROPIC_API_KEY` | | <img src=".github/assets/logos/cursor.png" width="20" valign="middle"/>  **Cursor Agent** | `agent` | Cursor subscription | | <img src=".github/assets/logos/copilot.png" width="20" valign="middle"/>  **GitHub Copilot** | `copilot` | Copilot subscription | | <img src=".github/assets/logos/codex.svg" width="20" valign="middle"/>  **Codex** (OpenAI) | `codex` | ChatGPT auth or `OPENAI_API_KEY` | | <img src=".github/assets/logos/gemini.svg" width="20" valign="middle"/>  **Gemini** (Google) | native | Google auth | | <img src=".github/assets/logos/goose.png" width="20" valign="middle"/>  **Goose** (Block) | `goose acp` | provider key | | <img src=".github/assets/logos/qwen.svg" width="20" valign="middle"/>  **Qwen Code** | `qwen` | Qwen auth | | <img src=".github/assets/logos/opencode.svg" width="20" valign="middle"/>  **OpenCode** | `opencode` | provider key | | <img src=".github/assets/logos/kimi.svg" width="20" valign="middle"/>  **Kimi** (Moonshot) | `kimi` | Kimi login | Plus **Factory Droid**, **Augment**, **CodeBuddy**, **Qoder**, **Kiro**, **Mistral Vibe**, **Snow**, and any custom ACP agent. 18 ACP CLI agents in all, plus native Gemini and the bundled Fuigo engine. **Engine providers** (Fuigo): one [Flux Router](https://fluxrouter.ai) key reaches every model Flux routes. To use a Claude subscription with no key, run the Claude Code backend and sign in with the `claude` CLI. ## How it works Wayland runs a four-step loop on every turn: - **Perceives** your request and the state of your files, project, and memory. - **Reasons** with the best model for the task. A read-only Plan mode can write a structured plan before anything is touched. - **Acts** through built-in tools (Read, Write, Edit, Bash, Grep, Glob, Spawn) and connectors for Git, databases, and the web, inside the engine's workspace sandbox on macOS and Linux. - **Evolves**: a loop mutates and scores your skill prompts against an eval harness, keeping a variant only when it beats both the running best and the parent it came from. Winners persist across runs, so the next run starts from the last one's best. Nothing reaches your live library until you promote it. **Fuigo engine.** One self-contained Rust binary, no Node or Python runtime to install. Fuigo is a Ferrox Labs fork of [xai-org/grok-build](https://github.com/xai-org/grok-build) (Apache-2.0), bundled per platform and driven over ACP on stdio. It ships the built-in tools, the MCP client, memory, sub-agents, and a workspace sandbox on macOS and Linux. The same engine powers the standalone `fuigo` CLI and the desktop app: one codebase, two surfaces. **Flux routing (optional).** Route a backend's traffic through Flux Router to send each task to the best-fit specialist across same-class models and run multi-AI cross-audit, lifting quality while cutting wasted tokens. Opt-in, bring your own key, off by default. ## Build from source Requirements: [Bun](https://bun.sh) 1.3.x, Node 22 to 24, and your platform toolchain for native modules. ```bash git clone https://github.com/ferroxlabs/wayland.git cd wayland bun install # Run the desktop app in dev bun run start # Tests, lint, typecheck bun run test bun run lint bun run typecheck ``` Package installers with electron-builder: ```bash bun run build # macOS arm64 + x64 (default) bun run dist:win # Windows bun run dist:linux # Linux AppImage + deb + rpm ``` ### Standalone Fuigo CLI The engine ships on npm as `fuigo`. The launcher pulls only the binary matching your machine. ```bash npm i -g fuigo # Or run it with no install npx fuigo "Read Cargo.toml and explain the dependencies" ``` The CLI updates with `npm update -g fuigo`, independent of the desktop app, which always runs the version it bundled. ## Configuration and keys Wayland runs on your provider credentials. There is no required Wayland-hosted backend to chat. | What | Where | Notes | | --------------- | ------------------------------- | ---------------------------------------------------------------------------------------------------- | | Anthropic key | `ANTHROPIC_API_KEY` or in-app | For a Claude subscription with no key, use the Claude Code backend and sign in with the `claude` CLI | | OpenAI key | `OPENAI_API_KEY` or in-app | Covers OpenAI-compatible endpoints (DeepSeek, Ollama, and more) | | Other providers | in-app | AWS Bedrock, Google Vertex, per-CLI auth for each ACP backend | | Flux Router | `FLUX_API_KEY` (`sk-flux-...`) | Optional. Only needed if you route through Flux | | Constitution | `~/.wayland/CONSTITUTION.md` | An editable rulebook prepended to every turn, with per-specialist overrides | | Data and memory | SQLite under your OS config dir | Your files, chats, and memory stay on disk | The desktop app hands Fuigo your Flux Router key per spawn (`FUIGO_API_KEY`); the standalone CLI reads its own `config.toml`. ## FAQ **Are my keys and data private?** Yes. Keys are stored in the OS keychain and data lives in SQLite on your disk. There is no Wayland account and no Wayland server: when you send a prompt it goes to the provider you configured, on your key, and nowhere else. The engine's outbound traffic passes an egress gate that is enabled by default and classifies every request before it leaves. "Air-gapped" would be the wrong word for a tool you point at a cloud model, and it is the provider you chose that sees your prompt. **Can I run fully offline?** Yes. Point the engine at a local Ollama model and Wayland runs with no network at all. Voice dictation runs offline with a bundled Whisper model, and Wayland can read replies back to you with voice output, so you can work hands-free. **Do I need a Wayland account or subscription?** No. Bring your own provider keys, or use the Claude Code backend with your Claude subscription (sign in with the `claude` CLI). There is no paywall to do work. **My local models are not showing up.** Make sure Ollama is running and reachable on its default port, then refresh the model list in Settings. **Is it really open source?** Yes, both parts. The desktop app is GNU AGPL-3.0 and the Fuigo engine is Apache-2.0, so you can embed the engine anywhere. ## Contributing Contributions are welcome. Read [CONTRIBUTING.md](./CONTRIBUTING.md) before opening a PR. In short: Bun toolchain, strict TypeScript, Arco Design components, run `bun run test` and `bun run lint` before you commit, and keep commits in the `type(scope): subject` format. Do not add AI authorship signatures to commits. ## License & the Wayland name Wayland is **real open source**. This desktop app is [GNU AGPL-3.0](./LICENSE); the engine, [Fuigo](https://www.npmjs.com/package/fuigo), is Apache-2.0. The split is deliberate: a permissive engine so anyone can embed it, copyleft on the app so the GUI stays open. Run it, self-host it, modify it, fork it, and build commercial services around it. The only catch AGPL adds: a networked service built on the app must publish its source under the same terms. Contributions are under a light [CLA](./CONTRIBUTING.md); third-party attributions live in [notices/](./notices/). **Where it came from.** This app originates in part from [AionUi](https://github.com/iOfficeAI/AionUi) (Apache-2.0) — parts of the Electron main process, the IPC bridge, renderer scaffolding, ACP integration and MCP services — and has since diverged substantially. The engine, Fuigo, is a Ferrox Labs fork of [xai-org/grok-build](https://github.com/xai-org/grok-build) (Apache-2.0), with the upstream copyright and NOTICE preserved in the bundled `notices/`. Full attributions: [notices/THIRD-PARTY-NOTICES.md](./notices/THIRD-PARTY-NOTICES.md). **Who builds this.** Ferrox Labs is a small team, and most of us are part-time. We use Wayland to build Wayland, which is why the shipped surface is larger than a headcount would suggest — and why the honest answer to "has this been audited?" is on the [evidence page](https://getwayland.com/built-on), not buried. A hosted **Wayland Pro** with expanded capabilities is on the way. The core you self-host stays complete and free, never crippled to sell you the hosted one. The **code** is AGPL; the **name and logo** are trademarks. Fork freely, just give your fork its own name. You can always say it's "built on Wayland" or "compatible with Wayland." Full policy: [TRADEMARK.md](./TRADEMARK.md). Wayland is named after Wayland the Smith, the master craftsman of Norse and Germanic legend, the one who could forge anything. A tip of the hat to the Wayland display server protocol; different project, no affiliation. ### Approval modes for scheduled tasks and workflows New scheduled tasks and workflow runs use guarded approval defaults: Auto-Accept Edits for Gemini, Accept Edits for Claude, and Auto Edit for Codex. Other backends use their normal mode without Desktop enabling blanket auto-approval. Choose the approval mode in the scheduled-task editor or the workflow launch dialog; a global preference from another conversation does not authorize a new workflow to run in Autopilot. Fuigo runs scheduled turns unattended through the same ACP permission model as a chat; approval mode does not widen workspace filesystem boundaries. Select Autopilot or Full Auto explicitly for a run only when that is the intended policy. Existing tasks retain supported stored modes when edited; review that field when updating an older task. Unrecognized mode values, including the old generated `bypassPermissions` token, use ask-first rather than enabling full-auto. A scheduled turn is not dispatched when its requested settings still fail after a fresh-session retry. Starting a fresh workflow uses the displayed approval choice; resuming an existing workflow returns to that conversation’s current policy.

wayland

内容

连接信息

猜你喜欢

wayland

微信扫一扫分享

需要认证

内容

连接信息

猜你喜欢