yothere vs cloud coding agents

The architectural difference is the whole story. Cloud coding agents (OpenAI’s Codex cloud tasks, Claude Code on the web, Devin, and the background agents in AI editors) run your tasks on the provider’s compute: your repository is cloned or mirrored into a sandbox they operate, the agent works there, and you review from a browser. yothere makes the opposite bet: we host the cockpit, the inbox, and voice, and run zero agent compute. The agents run on your machine, through your own Claude Code or Codex CLI login, on the subscription you already pay for.

Neither bet is universally right. Here is the honest map.

What crosses the wire

With a cloud agent, your code is on their infrastructure for the duration of the work; that is what makes it work with your laptop closed.

With yothere, your files are never uploaded. What crosses to the hosted control plane is the thread: the task you dispatched, its status, the agent’s own summaries, questions and results, and the cost ledger. If an agent quotes a line of your code in an answer, that quote is stored, because it is what the cockpit shows you. The repo itself stays on your box. The full picture is on the trust page.

The cost model

Cloud agents meter their compute: usage-based pricing, credits, or a seat that prices their infrastructure in. yothere cannot meter what it does not run. The plan is $20/mo flat after a 7-day free trial, and inference happens on the Claude or ChatGPT subscription you already have. Runaway-cost protection is still built in: daily and per-thread caps, on by default, and a capped thread comes back to your inbox as a question instead of freezing.

The honest trade

The one thing cloud agents genuinely have over yothere: their machines do not close. yothere’s work advances only while your paired machine is awake and connected. If your laptop sleeps, dispatched threads queue on the control plane and drain the moment it reconnects. Nothing is lost, and nothing advances either. That is the deal you make for your code never leaving your machine and your inference never being resold to you at a markup. We say the same thing in our FAQ and our llms.txt, because it is the single most important qualifier of the product.

Beyond code

Most cloud coding agents are repo-scoped by design: the unit of work is a pull request. A yothere hail can be a code task bound to a repo (each thread works in its own git worktree, on its own branch), but also research, drafting, review, or planning: “scan our top 3 competitors”, “draft replies for the support inbox”. One inbox for all of it.

Side by side

	Cloud coding agents	yothere
Where agents run	Provider’s sandbox	Your machine
Your repo	Cloned to their infra for the task	Never leaves your box; yothere never clones
Works with laptop closed	Yes	No; work queues until your machine is back
Inference billing	Their meter (usage, credits, or seat)	Your existing Claude/ChatGPT subscription; no metered bill
Task types	Mostly repo-scoped	Code, research, drafting, review
Voice	Generally none	Hosted, 120 min/day, hands-free loop
Who can publish code	The agent opens PRs from their infra	An agent can never git push; you press Ship it, and the push runs as you, outside the agent’s turn
Price	Varies by usage/seat	$20/mo flat after 7-day free trial

When a cloud agent is the better fit

If you need work to continue while every machine you own is off, that is their feature and not ours. Same if you cannot leave a Mac or Linux box awake, or your team standardizes on a managed sandbox for policy reasons. We would rather you pick the right architecture than squint at this page.

FAQ

Does my code get uploaded to yothere?

No. The agents run on your machine in the directory you point them at. The hosted control plane stores the thread (task, status, summaries, questions, results, cost), not your files. Anything an agent writes back into a thread is stored, because it is what the cockpit shows you.

Does yothere keep working when my laptop is closed?

No, and we will not pretend otherwise. Dispatched threads queue on the hosted control plane and drain the moment your machine reconnects. Nothing is lost. If closed-laptop continuity is your hard requirement, a cloud agent is the better fit.

Can yothere host the compute for me?

No. Bring-your-own-compute is the design, not a missing feature: it is why your code stays home and why there is no inference meter. You bring a Mac or Linux machine you can leave open (Windows via WSL2) and your own agent subscription.

Which agents does yothere run?

Claude Code (the default), the Codex CLI (a ChatGPT subscription is enough), and the open-source OpenClaw CLI. Anything that speaks the Brain Protocol can be wired in as a remote brain. See how it works.