An AI employee that actually does the work.
Describe a task in plain English. myrt opens a real browser, clicks, types, reads and researches — and you watch every step live in a viewport, down to the last cent of what it costs.
Free tier includes $0.50 of daily agent compute. Your card is never charged without your action.
Three steps. No prompt-engineering degree required.
Describe the task
“Check these three flight sites and find the cheapest Thursday flight from Jakarta to Tokyo.” Plain English. The agent plans its own steps.
Watch it work
A live viewport streams what the agent’s browser sees. Every navigation, click and keystroke appears as it happens — with a running cost meter.
Get the result
A direct answer with the work shown. If a site needs your login, the agent asks and you sign in yourself inside its browser — your password never touches the model.
Built for real work, not demos.
The boring parts done right — that’s what makes an AI worker trustworthy.
Transparent by default
Every run shows its steps, its browser viewport and its exact cost. No hidden token markups — a research run typically costs $0.01–$0.05.
Isolation per tenant
Every user’s agent runs as its own OS user with its own workspace, browser profile and filesystem — enforced by the kernel, not by convention.
Human-in-the-loop & accountable
Login walls stop the agent and ask you to sign in yourself. A Report button on every finished run sends problems straight to the operator with context attached.
Open, self-hosted core
myrt is a single 140 MB binary — runtime, agent and dashboard included. Run the whole platform yourself, or use the cloud. Your data stays yours either way.
curl -fsSL https://myrt.cloud/install.sh | sudo sh
install script · binary · self-host guide
Scales with the work, not the seat count.
Daily agent-compute budgets. No per-seat fees, no minimums, cancel any time.
Free
- Real agent runs, not a demo sandbox
- Live viewport + full step log
- Browser automation included
- No card required
Plus
- Bigger daily allowance
- Full wasm surface
- More API keys
- Same kernel isolation
Pro
- Priority agent scheduling
- Longer session memory
- Bigger daily budget
- Cancel any time
Enterprise
- Custom limits + SLA
- Self-host or dedicated box
- Audit exports + RBAC
- Onboarding + support
Full plan limits and checkout live on the pricing page; the API contract is in the docs.
Questions people actually ask.
What can the agent actually do?
Anything a browser and a shell can: research across multiple sites, fill forms, extract structured data, monitor pages for changes, summarize long documents, run small scripts on the results. Each run shows the steps it took, so you can verify the work instead of trusting a black box.
How does pricing really work?
You get a daily compute budget ($0.50 on the free tier). Each run costs its true model cost — typically $0.01–$0.05 — and the live meter shows it as it accrues. Nothing is charged automatically; when the daily budget is spent, new runs wait for the next day (or an upgrade), and a run in progress is never cut off mid-task.
What happens when a site requires my login?
The agent detects the login wall and asks you for help. A live browser window opens that only you control; you sign in with your own credentials while the agent waits. The agent then continues with the authenticated session — it never sees what you typed.
Is my data isolated from other users?
Yes — at the OS level. Each tenant’s agents run as a separate system user with a separate workspace, browser profile and process namespace. This is enforced by the kernel; a compromised agent process still cannot reach other tenants’ files.
Can I run myrt myself, without the cloud?
Yes. The whole platform is one self-hosted binary for Linux x86-64: curl -fsSL https://myrt.cloud/install.sh | sudo sh verifies the SHA-256 and installs it (the raw binary and its checksum are there too). The myrt-connect browser extension pairs a self-hosted instance with your browser for live approvals and viewport streaming.