Short answers to the questions we get most. For the how, follow the links into the docs.
Basics
What is yothere?
yothere runs up to five agent turns at once and gives you one inbox instead of five terminal tabs. You hail a task by voice or by typing, the agents work on their own on your machine, and yothere brings you back only when a thread genuinely needs a human. We host the cockpit, the inbox, and the job queue; your own machine runs the agents. See Core concepts and How it works.
What do I need to bring?
Four things: a Mac or Linux machine you can leave open (Windows works via WSL2), Python 3.11 to 3.13, pipx, and your own Claude Code or Codex CLI login (an existing Claude or ChatGPT subscription). yothere never holds your model credentials and never bills you for inference. See Onboarding.
What happens when my laptop is asleep or closed?
Nothing advances, and nothing is lost. The agents run on your machine, so when it is off, dispatched tasks queue on the hosted control plane and wait. The moment your machine reconnects, the “no machine connected” banner clears and the queue drains. This is by design: your compute is your machine, so leaving the laptop open is part of the deal.
How is this different from just running Claude Code?
Claude Code is one agent you drive turn by turn: you type, you wait, you read, you type. yothere advances up to five turns in parallel and routes your attention across them, so you delegate and walk away and it brings you back only when one thread is blocked or done. Typed questions even run on their own lane, on top of the five, so a reply never queues behind the work. Claude Code (or Codex) is still what does the actual work underneath. Full comparison: yothere vs Claude Code.
Is there a hosted version?
That is the product. The cockpit is live at app.yothere.ai. Signup is open, and your first 7 days are free with no credit card. After the free week, the $20/mo Founding plan keeps the workspace running. See Pricing.
Cost and keys
What does it cost to run?
The agents run on the coding-agent subscription you already pay for, so there is no metered inference bill from us. To keep a runaway fleet from surprising you, the runner ships with built-in cost caps: $10 per day across the fleet and $10 per thread, both configurable. A capped thread comes back to the Inbox as a card you can answer (continue it, raise the cap, or stop it). See Configuration.
What does the hosted plan cost?
Your first 7 days are free, no credit card required. After that it is $20 a month, the Founding price, locked in for as long as you stay, even after we raise it. It covers the cockpit at app.yothere.ai, the inbox, and hosted voice (120 minutes a day, resets daily). Your agents still run on your own machine, on the Claude or ChatGPT subscription you already have, so there is no metered inference bill from us on top. Teams, more seats, or higher limits are available by request. See Pricing.
Do I need API keys?
No. The agents run on your existing Claude Code or Codex login, so we never hold your model credentials and never pay for your inference. Voice runs on our key: the control plane mints a short-lived token per call, and you bring nothing. The bundled demo (yothere init --demo) also runs end to end with zero keys.
Voice, brains, platforms
How does voice work?
Voice is hosted. Click Connect in the cockpit and talk to your fleet in the browser. There is no phone number, no Twilio, and no key of your own. The plan includes 120 voice minutes a day (the allowance resets every day), so a forgotten call cannot run away. Ask “what’s waiting on me?” and it reads your inbox back, hand off a new task, answer a blocked thread, say “send it” to approve a draft (it records a real approval), or pick a finished thread back up. Tap the mic button (or say “mute me”) to mute yourself; the button is the way back, because a muted mic cannot carry the word “unmute”.
The call runs on our media worker, not on your machine. The transcript streams into the cockpit live while the call runs, and the call content is deleted from our servers the moment the call ends: we do not keep it after the call (1.23.0 privacy). The audio is not recorded. While a call is live, treat what you say the way you treat typing into the cockpit. See Privacy. English, or neutral Latin American Spanish, is a setting in /settings.
Can yothere write code?
Yes, and that lane shipped. Register a git repo you already have (yothere workspace add site ~/code/my-site), then hail a change: yothere spawn --repo site "fix the layout shift on the pricing page". yothere never clones. The thread runs in its own git worktree on its own branch, so it cannot touch your working tree or another thread’s, and it commits its work and ends with the diff, for you to read. Press Ship it and yothere pushes that branch and opens the PR as you (see below). Full surface in the CLI reference, a worked recipe in Tutorials.
Can an agent push to my repo?
No, and no approval can change that. git push, gh api, and the mutating gh subcommands (create, comment, merge, and the rest) are on a never-grant list inside the send-gate: the approval path cannot mint a grant for them, no matter what you click. It is the strongest rule in the gate, and it is still a rule in the gate rather than a sandbox: the trust page names the two paths that sit outside it. The agent is told this up front, so it does not burn its turn hunting for a workaround that does not exist.
yothere itself does push, though, and only because you asked it to. A code thread ends with the diff it wrote. You read it, press Ship it, and the branch is pushed and the PR opened as you, with the git and gh login already on your machine. That rail is deterministic git with no model in the loop, and it runs outside the worker turn, so the thing that wrote the code is not the thing that publishes it. yothere holds no credential of yours: if gh is not authenticated, it says so rather than asking for a token.
Can it schedule a task, or run one every morning?
Not yet, and this is the one thing on the docs that is genuinely not built. There is no scheduler. A hail, or a preset, runs once, immediately. “Every morning at 7, scan the news” spawns one thread that runs now. Scheduling is not built yet.
Which agents or brains does it support?
Three local coding agents ship today: the Claude Code CLI, the Codex CLI, and the open-source OpenClaw CLI, so a Claude subscription, a ChatGPT subscription, or OpenClaw’s own model provider is enough to run yothere. Claude Code and Codex are peers, with the same containment, the same approve-send grant, and the same caps; OpenClaw is contained to the observe/workspace tiers. Beyond those, anything that speaks the Brain Protocol works: a remote brain reachable over ws:// or wss://, which you can implement against your own stack. See How it works and Configuration.
macOS or Linux?
Both. The always-on service installs as a launchd agent on macOS and as a systemd user unit on Linux, via yothere service install. Windows is supported via WSL2, which takes the Linux path. Python 3.11 to 3.13 (3.14 is not supported). See Onboarding.
Privacy and safety
Does my data leave my machine?
Your files are never uploaded. The agents run on your machine, in the directory you point YOTHERE_WORK_DIR at, on your own agent CLI and your own subscription: we never copy your repo or your work directory anywhere.
What does cross to the hosted control plane is the thread: the task you dispatched, its status, the agent’s own summary, questions and results, its live activity line (the tool it is running and the file or command in it), and the cost ledger. So if the agent quotes your code back in an answer, that quote is in our database, because it is what the cockpit shows you. Assume anything the agent writes back is stored (a live voice transcript is the exception: it crosses only while the call runs and is cleared when it ends). We never hold your model credentials, and the device credential written by yothere login carries no compute keys. See the Trust page and Privacy for the full picture.
What will yothere never do?
Decide that your code is ready. git push and gh are permanently excluded from what any approval can grant an agent, so a worker turn can never publish: yothere pushes only when you press Ship it on a diff you have read, and then it runs as you, outside the turn. The send-gate also holds every other outward action (mail, messages, posts) until you approve it. Be clear-eyed about the boundary, though: the gate is a guardrail, not a sandbox. It inspects shell and tool calls, so what an agent reads and its plain GET requests are not contained. Pair a machine whose reachable data you are comfortable handing to the control plane. See the Trust page.
Is it open source?
The package is published on public PyPI under the MIT license (pipx install yothere). The source repository is private during the beta, so a source install is not the public path; install from PyPI, and audit the sdist if you want to read what runs.
How do I get access?
yothere is in open beta. Sign up. Your first 7 days are free, no credit card. Then the Founding plan is $20/mo. Questions? Email [email protected]. See Pricing.