yothere

Introduction

yothere runs up to five agent turns at once on the Claude or ChatGPT subscription you already pay for, and gives you one inbox instead of five terminal tabs.

yothere runs up to five agent turns at once on the Claude or ChatGPT subscription you already pay for, and gives you one inbox instead of five terminal tabs.

You hail a task, by typing or out loud. The agents work on their own, on your own machine. They come back to you only when one of them genuinely needs a human.

The name is the interaction. You say “yo, you there?” and your fleet answers “yohere, on it.”

Start hereInstallGet yothere on your machine and pair it to the cockpit. 5 minutesQuickstartYour first hail, end to end. The ideasCore conceptsThreads, brains, the send-gate, the code lane. Under the hoodHow it worksThe core loop, and what runs where.

Before your first command

yothere is a cockpit, not a model. It never holds your model credentials and never bills you for inference. You bring four things:

You bring Why
A Mac or Linux machine you can leave open The agents run here, not in our cloud. Windows works through WSL2.
Python 3.11, 3.12, or 3.13 3.14 is not supported and yothere doctor fails it.
pipx Installs the CLI into its own environment, persistently.
Your own Claude Code or Codex CLI login The turn runs as you, on the subscription you already pay for.
Leave the laptop open. The hosted control plane runs zero agent compute. Every turn runs on your paired machine. While that machine is asleep or closed, dispatched threads sit in the queue and drain the moment it reconnects. Nothing is lost, and nothing advances either.

The loop

  1. Hail. One line, typed into the cockpit, spoken on a voice call, or run as yothere spawn "scan our top 3 competitors" from the CLI.
  2. The fleet works. The cockpit queues the thread. Your paired machine leases it and runs the turn locally, on your Claude Code or Codex login. Up to five work turns advance in parallel by default, and running out of steps is not a failure: the thread takes another turn.
  3. One ping, or none. Only what genuinely needs your call reaches the Inbox, and a push lands on your phone when it does. Reply from the card, and the fleet moves again.

Four kinds of card can reach you

  • A question. It hit a real fork and needs your call.
  • A draft to send. It wrote the email. You press Send.
  • A plan to approve. It wrote the approach before doing the work. Approve, or send it back.
  • A result to look at. It finished. Glance, confirm, done.

Everything else advances silently. Most of the time the Inbox is empty. That is the product working.

Your reply never queues behind the work: typed questions run on their own pool, on top of the five work turns, so you can interrupt one agent while the other four keep going.

What you get

  • The cockpit at app.yothere.ai/overview, with three lanes: an inbox of what needs a human, a working lane of what is in flight (and what each turn is doing right now), and a done pile.
  • Hosted voice. Click Connect in the cockpit and talk to your fleet in the browser. No phone number, no Twilio, no API key of your own, and 120 minutes a day (resets daily). The transcript streams into the cockpit live during the call, and the call content is not kept on our servers after the call ends. See Privacy.
  • The code lane. Register a repo you already have with yothere workspace add, then hail yothere spawn --repo <alias> "fix the layout shift on the pricing page". The thread works in its own git worktree, commits to a branch, and hands it back for you to read. No approval opens a push or a merge. See Core concepts.
  • Push to your phone. Install the cockpit as an app, and a push deep-links to the one thread that blocked.
  • A send-gate and cost caps on your machine. Outward actions are held until you approve them.

What holds the blast radius, honestly

Worker turns run without per-action prompts, so the guardrails matter and we state them plainly:

  • The send-gate. Outward actions (sends, posts, pushes) are held on your machine until you approve them.
  • No approval opens a push. git push and gh can never be granted, by any approval, ever. There is no button that publishes.
  • Cost caps, per thread and per day, on by default. Voice has its own daily allowance: 120 minutes, resetting every day.

The send-gate is a guardrail, not a sandbox. It inspects shell commands and MCP tool calls, and nothing else: what an agent reads, what it writes to disk, and its outbound GET requests are not contained. See Trust for the full boundary.

Not built yet

Coming soon Recurring tasks. A hail runs once, immediately. "Every morning at 7, scan the news" is not a thing yet, and there is no scheduler in the product today. It is not built yet.

It is not on sale, and it has no date. A shipped date we miss is the same lie in a slower voice.

Where to go next

yothere is an open beta. The cockpit, the inbox, and the job queue are hosted at app.yothere.ai. Sign up. Your first 7 days are free, no credit card, then the Founding plan is $20/mo. You pair your own machine, so your own agent subscription runs the turns, and nothing goes to us for inference. See Pricing.