the terminal is the new classroom
Building with AI agents is a practice, not a syllabus — you learn it the way you learn to drive. Tandem puts you in a live terminal with someone who already ships with Claude Code: watch them work, then take the wheel while they’re beside you.
Software skill has always transferred by watching, then doing with someone beside you. AI agents made the expert’s screen the scarce resource again — Tandem shares it.
Your instructor drives their agent on a real task while you watch the terminal live. The failures are the lesson — and nothing is edited out.
Control passes to you mid-session. You phrase the prompts; the instructor watches, corrects, and takes back the wheel when things go sideways.
Run your own session. The transcript captures every prompt you wrote, so your instructor can review how you think — not just what you shipped.
No slides, no sandboxes, no edited screencasts — a real agent, on a real machine, working on a real codebase, shared live.
The instructor’s agent runs on THEIR machine — Claude Code, Codex, any CLI. Tandem streams the real terminal to your browser: every prompt, every correction, every recovery, unedited.
One driver at a time, by design. The instructor demonstrates, then hands you the keyboard — you drive the agent yourself while they watch and step in when you wedge.
Chat sits next to the session, so questions never interrupt the flow. Drop files and screenshots straight into the agent’s working folder.
Sessions record automatically. Browse what was typed — the exact prompts, the wrong turns, the fixes — any time after the session ends.
Git branch, dirty state, ahead/behind — visible to everyone live. You learn on real projects with real stakes, not toy exercises.
The bridge makes an outbound-only connection from the host’s machine. API keys, dotfiles and repos never leave home.
One command from the dashboard puts the host’s terminal in a room. Outbound-only — no ports, no VPN, no firewall fights.
tandem.duosh.dev/app → + New terminalSend the room URL. Students sign in with GitHub and are watching the session in seconds — nothing to install.
relay.duosh.dev/?room=…Chat, pass control, drop in files. When the session ends, the transcript is already in the library.
📜 transcript savedLearning is always free — only hosts pay. Invite someone new and you both get two weeks of Pair free.
Learn by watching
1:1 teaching & mentoring
Cohorts & classrooms
Billing launches with the beta — early users keep free access until it does.
Working with AI agents is judgment, not syllabus — when to re-prompt, when to interrupt, when to revert. Courses go stale at every model release; an edited video hides the recoveries, which are the actual lesson. Watching a live session — then driving one — is how the skill transfers.
The agent runs on the host’s machine; only the terminal stream (what you’d see on screen) relays through the hub, encrypted in transit. API keys, dotfiles and repos stay local.
Anything that runs in a terminal. It’s tuned for AI coding agents — Claude Code first — but a plain shell, vim, or any REPL streams just as well.
No. Joining a session is always free with an account — only the host needs Pair or Team. And when you invite someone new, you both get two weeks of Pair free.
Absolutely. The teaching flow IS the pairing flow: live shared terminal, control handoff, chat, transcripts. Pair is pair programming; Team is your team’s pairing home.
They’re private to you (or your cohort) by default. Browse the transcript, share deliberately, or delete — deletion is real deletion.
🚲 Tandem — a duosh.dev project · Your terminal stays yours.