Docs

Getting started

Install Wilow, create an account, connect your AI engine, and run your first project.

1. Install Wilow

  • Windows (recommended first install): download the installer from the homepage. The installer is currently unsigned, so Windows SmartScreen may warn on first run — choose More info → Run anyway. The desktop app bundles the engine and auto-updates itself.
  • Web / phone: open wilow.team/app — a full PWA you can add to your home screen, with push notifications. Settings → Notifications decides what each device hears about beyond Wilow's replies: team chats (on by default), History (off by default), and the projects you silenced from their History card or team chat.
  • Android: a signed APK is available on the homepage (sideload; enable “install from unknown sources”).

The desktop is the home of your account

Your computer runs the engine — projects live in local folders and all builds execute locally. Phones and browsers are windows into it. Install the desktop app first, then add devices.

2. Create your account

  1. Open the desktop app (or the web app) and press Continue with Google, or choose Register for an email + password account. On the desktop, Continue with Google is one Google page that both creates your account and lets your agents use Gmail, Docs, Sheets, Calendar, Drive and Analytics right away (see step 4 for what that grant covers). The Android app uses email + password for now.
  2. Passwords must be at least 8 characters and are checked against known-breach lists before being accepted. You'll receive a 6-digit verification code by email; enter it to activate the account.
  3. On the desktop, signing in provisions this computer as your account's engine and generates your encryption key locally.
  4. To add your phone (or tablet): open the web app on it, sign in, and scan the QR code shown by the desktop app — your first time, click Pair a device at the bottom of the sidebar; after that, pairing lives in Settings → Devices → Pair new remote control. You can pair as many devices as you like, and each one asks for a name as it finishes setup. The QR carries only the encryption key — scanning it is what lets that device read your chats. Every paired device is listed under Settings → Devices, where it can be renamed or revoked later.

3. Connect your AI engine

Wilow's agents run on an AI engine you choose during setup: OpenCode (the default) or Claude. Pick OpenCode and onboarding immediately asks which model provider it should run through — OpenRouter (the default) — one API for many models, pay-per-token, your own key; the desktop app connects it in one click (an OAuth flow; the key is written to a local file on your computer only), and costs appear on your OpenRouter dashboard — or Fireworks, which runs your agents on open-weight models from a pasted API key. Choose Claude instead and you sign in with your own Claude subscription (Pro/Max); Wilow installs Claude Code locally and drives it, so your Claude account and its usage limits apply. Both engines use the same seven model aliases (frontier, coder, reasoning, design, statistical, default, cheap) — on Claude each alias maps to an Anthropic model you can retarget in Settings → Models (Sonnet 5 powers the working aliases by default; Fable, Opus and Haiku are available too). Change backends later in Settings → Models (new projects) or per project in each project's settings — switching keeps your chat history and project files.

4. Connect your accounts (optional but recommended)

In Settings → Connections you can link GitHub (repos and PRs), Vercel / Netlify / Firebase Hosting (deploys), Supabase / Neon (databases), Google Workspace, Stripe, and email. Two rules apply to all of them:

  • Credentials stay local — tokens/keys are stored in a file on your computer and are never uploaded.
  • Outward actions are approval-gated — anything that leaves your machine toward other people (sending email, charging a card) pauses and asks you first.

Google Workspace has two tiers. Connect with Google (one click, Wilow's own Google client) lets the agents send mail, create and edit Docs and Sheets, manage Calendar, read Analytics and work on Drive files they create or you hand them. Reading your inbox and browsing your existing Drive files are permissions Google only grants a shared client after a paid audit, so those come from Advanced: use your own Google project: a Google Cloud project of your own, once, so the agents run under your own Google quotas and no one else's. The in-app guide walks Google's current console page by page — Branding (with copy buttons for the exact values Google requires before it will publish an app), Audience (publish it — leaving it in "Testing" makes every sign-in fail with Error 403: access_denied), Clients (a Desktop app) — and tells you in advance about Google's "hasn't verified this app" page, which is normal for your own private app. Never upload a logo: it forces a full Google verification. Wilow installs the small launcher the Google tools need (uv) itself; no terminal is involved.

Everything on your computer that Wilow depends on is listed under Settings → Tools(desktop app): every command-line tool with its version and the account it is signed in as, every MCP server the engine can use with a Check button that starts it and confirms it answers, and every skill, command and agent. Connections is the short list of what Wilow needs per category; Tools is the full list. Each tool row has an Install button wherever Wilow can put it there itself — Git differs by platform: Windows gets it silently; on a Mac it comes with Apple's Command Line Tools, and Install opens Apple's installer. Copy report puts the whole list on the clipboard: when something does not work, paste it into your support message.

Three of those tools are mandatory: the engine that builds (OpenCode), Git and the GitHub CLI. Setup checks them on the computer that runs Wilow and installs the missing ones itself; it cannot finish while one is missing, and if one disappears later the chat shows a banner that opens Tools.

5. Your first project

  1. Tap New project in the sidebar, give it a name and a description of what you want built.
  2. Wilow creates a local folder, asks its discovery questions in the project chat, then builds continuously — commit by commit, with tests — reporting progress in the chat.
  3. Already have code? Pick Import an existing project in the same dialog and point Wilow at a local folder (it's copied into your projects directory) or a GitHub repository (it's cloned — no new repository is created). The project takes the folder's or repository's name, so the folder, the chat and the repository always match; if that name is already taken you pick another and Wilow renames the repository to match. Wilow reviews what exists and continues from there instead of starting from scratch.
  4. Each project chat has a Control Panel (recent activity, up-next backlog, links to the live deployment) and per-project settings, including Autopilot: scheduled maintenance cycles that fix bugs and propose improvements on a cadence you choose. Autopilot and each menu command (Production Scan, UI Check, Cybersecurity Scan, Add Agent, Merge PRs) can run on its own model alias — your chat always returns to its own model when they finish.
  5. When a project is finished (or a mistake), the chat's ⋯ menu offers Close — archives the chat and stops its work, reopenable any time — and Delete project, which permanently removes the chat and, as you choose in the confirmation, the project's local files, its GitHub repository and the hosting or database projects Wilow created for it (each verified before it is listed, all pre-selected). A second confirmation spells out exactly what goes and what is kept.

Install tools from the Marketplace

The sidebar's Marketplace tab lets you browse and install agents, skills, plugins, and full apps. Installs are hash-verified and consent-gated — see Installing from the Marketplace.

System requirements & network

  • Windows 10/11 or macOS (Apple Silicon or Intel) for the desktop app — the Mac build is signed and notarized by Apple, so it opens without a security prompt. Any modern browser for the web app.
  • Outbound HTTPS only — no inbound ports. The exact egress list for firewall allow-listing is in the Security model.
  • Developers publishing to the Marketplace also want Node.js ≥ 22.5 or the desktop app (which bundles the CLI).