How can we help?

Download it from vassalagent.com/download, open it, and drag Vassal Agent into Applications. If your Mac says it can't check the app, the download page shows the clicks under If your computer shows a warning. You only do it once.

Open System Settings, go to Privacy & Security, scroll down and press Open Anyway next to Vassal Agent, then confirm with your password. You only do this once.

Yes. Vassal Agent runs on Windows 10 and 11, and on Macs with macOS 12 or newer (Apple chip or Intel). The download page picks the right one for your computer.

Nothing, to start. At the end of setup, press Let it learn and it reads what is already yours, such as your profile and your website, in a few minutes. After that it keeps learning from everything you tell it and every job. See, change or remove anything it knows on Your Vassal.

Pick Claude or ChatGPT on the setup screen. It installs what's needed in the background and opens the sign-in page in your browser. Sign in, come back, done: no commands to type. ChatGPT runs through OpenAI's Codex, signed in with your ChatGPT account. Other AI tools that speak MCP, such as Cursor, Windsurf, VS Code and Claude Desktop, can connect too: open Connect in the app and press Connect next to it.

One you already pay for. On Claude, Claude Pro is enough for most people; the free Claude plan can't run the assistant. For ChatGPT you sign in with your ChatGPT account.

Only when a job needs one. If a job uses Gmail or LinkedIn and you are not signed in, it opens the site and asks you to sign in the usual way, then carries on. Your passwords stay on your computer and never reach us.

Type what you want in the assistant box in plain words, like you would to a person. For example: find 30 companies in Miami hiring a head of sales and list the founders with links.

Yes. You see the AI's own output as it works, next to the browser on the same screen. Run several jobs at once in separate tabs; each tab shows whether it is working, waiting for you or done, and how many sub-agents are running.

No. Before it sends, posts, submits, pays or deletes, it stops and asks. Nothing happens until you press Allow. Every step it takes is listed under Activity.

Yes. When it asks, press Allow for this job and it stops asking about that button on that site until the job is done, or Always allow here to stop asking for good (you can undo it in Settings, then Rules). Harmless buttons such as Search, Next and Save draft never ask. Paying and deleting always ask.

Each workflow runs two ways: by hand, or on a schedule you set with the clock button on its row. Its latest result shows on that row, ready to click open, copy or continue in the chat. Workflows run while the app is open, or in the background when Keep running when closed is on.

Press Show it once on the Workflows screen and do the job yourself in the browser. It becomes a workflow, with what you typed turned into questions for next time. Passwords are never recorded. The same button re-teaches a workflow that stopped working.

It notices browser work you do again and again, using the tools you actually use, and offers to take it over as a workflow. It suggests at most one a day, and nothing changes unless you say yes.

Under every table in the chat or on the Workflows tab: Copy as table, Save as spreadsheet, Send to Google Sheets (it opens a new sheet in the browser and fills it in) or Email it (a draft, never sent).

Yes. Settings, then Rules. Write them in plain words, like Never press "Send" on linkedin.com or Ask me before pressing "Pay". If your team shares rules, theirs come first and yours can only add to them.

Not yet. Team accounts, where one owner shares workflows and rules with everyone, are on the way. Today each person has their own Vassal.

From what you tell it, the moment you say it. From what is already yours: your profiles, your site, your documents. From the edits you make to its drafts, so it writes like you. From every job: it keeps the route it took, so the next time it goes straight there. Everything it knows is on Your Vassal, where you can change or remove any line.

Contact us and we set it up for your business.

No. Sign in to the app with the email we set you up with. There is no password: we email you a 6-digit code.

Install it on the new one and sign in with the same email, then press Allow on your old one: everything you taught it comes across. No old computer at hand? Use your recovery code from Settings, On every computer. Sign an old one out on your account.

They install themselves. The app downloads a new version in the background and switches to it the next time you quit and open it.

Open Connections and press Sign in under Your AI. If it keeps happening, sign out of your AI (Claude or ChatGPT) and sign in again when the app asks.

Open that site in the app's browser and sign in again. Some sites sign you out every few weeks for security.

Only if you turn them on in Settings, then Advanced. They are off unless you do. A report holds the error, the app and system version and a few log lines with links, file paths, emails and anything like a key removed. It never includes the pages you visit, what you ask the assistant or what you type.

Check its row on the Workflows tab and the chat for what it managed and why it stopped. Press Run again, or tell it what to do differently. It keeps what it learned.

Contact us. A person answers.