Runvault is an AI agent with its own computer and a sign-in to your tools. Describe a job once, like a revenue report or a morning error triage, and it does it now and again on schedule.
Sign in with Google. No card, no API key.
Demo: on Friday at 5:42 PM, Sam asks to be emailed last week's Stripe revenue by plan every Monday at 9. The agent pulls the invoices through the Stripe connector, shows a table and what changed, emails it through the Resend connector, and sets up the schedule once Sam approves the command. The weekend passes, and at 9:00 AM on Monday the next report is built and emailed without anyone asking.
Say what you need in plain English. Runvault signs in to your tools, writes and runs the code on its own computer and hands you the result. Ask for it every week, and it comes back every week.
It has its own Linux computer with Python, Node and the usual tools. It writes the
script, runs it, and keeps what it makes in /workspace for next time.
Add any MCP server by its address and sign in once. For any other HTTP API, store a key and Runvault adds it to the agent's requests.
"Every Monday at 9" or "weekdays at 8:30" becomes a schedule. It runs on time whether or not you're there. The result waits in the chat, or lands in your inbox through a connector like Resend.
Tell it once and it sticks across every chat. Save the prompts you reuse as skills and start a chat from one.
Reports, triage, a question about your data, the chore that spans three tools. Each of these is a first message: paste one into a new chat.
Every Monday at 9, email me last week's Stripe revenue by plan and what changed.
Each weekday at 8:30, go through new Sentry errors and file the real ones in Linear.
Which PostHog events do people who upgrade fire in their first week that others don't?
Turn this week's Notion meeting notes into Linear issues, one per action item.
Here's our signups CSV. Chart weekly churn and give me a one-page PDF.
Check our pricing page every morning and tell me if anything on it changes.
Handing work to an agent means giving it access. Your keys still never enter its computer: Runvault adds the key to a request on its way out, and only for the host you tied it to.
$ curl api.stripe.com/v1/invoices $ printenv | grep -i stripe (nothing)
host matches api.stripe.com + Authorization: Bearer sk_live_•••• other hosts get nothing
200 OK { "object": "list", "data": [ … 48 invoices ] }
The key isn't in the agent's environment, files or logs, so a bad prompt has nothing to steal.
Your Stripe key only ever goes to api.stripe.com. A request anywhere else goes out without it.
Each command, tool call and API request lands in your audit log with its host and result.
The agent says what it's about to do and waits. Approve it, reject it, or let the rest of the run go ahead.
An approval card in the Runvault chat: the agent asks to run python3 /workspace/reports/stripe_weekly.py and waits until you press Approve or Reject.
Credits cover the model and the sandbox. No separate OpenAI bill.
Everything on this page, with room to try it properly.
For an agent you lean on every day.
Sign in with Google and tell it what you need done. No card, no API key.