The workspace that learns how you work.

rterminal is an agent with full control of your computer. Do a job with it once; it learns the workflow and runs it for you from then on.

The first time, you walk it through. After that, it's one line.

Every job you run with rterminal teaches it. What worked, and how you corrected it, is written into a skill: a plain file you can read, edit, and keep.

First run: you shape the work as it happens.

Every week after: one command.

Plain files, not a black box

Every skill is a readable file. Open it, edit it, write your own from scratch.

Corrections stick

Say it once. The skill updates, and every run after remembers.

Yours on every machine

Skills sync with your account. Sign in anywhere and your whole setup is there.

Missing a capability? It builds it.

When a job needs a tool that doesn't exist yet, rterminal designs one, writes it, and proves it against cases with known answers before trusting it on yours.

  • New tools join your kit: named, editable, ready for the next job
  • Tested against your real cases before they're relied on
  • Built once, used forever, on every machine you sign in on

Most software ships with a fixed feature list. rterminal grows one, shaped by your work.

One agent. Your whole computer.

rterminal works the machine the way you do, and shows every step in the trace as it happens.

Browser and web

Signs in, navigates, fills forms, downloads, and gathers what you need from the sites you already use.

Files and documents

Reads PDFs and scans, drafts and edits documents and spreadsheets, and reorganizes whole folders in place.

Terminal and tools

Runs scripts and command-line tools, installs what it needs, and chains them into complete pipelines.

Hours-long workflows

Carries a many-step job across a whole afternoon, pausing at the points where you'd want a say.

You set the limits per session: read-only, ask before acting, or full access. Every action is written to the trace, so you can always see exactly what it did.

Runs on the agent you already trust.

Pick a backend per session. The workspace, your skills, and your tools work the same on all of them.

Claude Code

Anthropic's coding agent, reasoning and tool calls in the trace.

Gemini

Google's CLI, normalized to the same reply and approval model.

Codex

OpenAI's agent, from sandboxed to full access.

OpenCode

The open-source agent, on the same surface.

Built on the open-source VS Code core. Runs on macOS and Linux.

Try it on your real work.

rterminal is in a private early access beta — free while it lasts, no card required. Sign up with your email and the early access token you've been given. At the University of Oxford? Use your university email on the Oxford path (an early access token is needed there too during the beta).