Moshi vs Anthropic Remote Control

Onyx Okay, this is actually a clean launch. If your coding agent is somewhere else and you’re staring at a phone, the question is whether you want Claude on a leash or your whole machine in your hand.

Echo Yeah. And that distinction matters more than the marketing copy. Remote Control is Anthropic’s own path into Claude from another device. Moshi is a mobile terminal that happens to be agent-aware.

Onyx That’s the part I like. It’s not just “remote AI.” It’s, do you want to reach one agent, or do you want the shell, logs, tests, and the rest of the workflow too? That is such an Exploring Next take, by the way.

Echo Mm-hm.

Onyx Also, small vibe check, how’s your week been? Mine is weirdly fine, which feels suspiciously unlike us. I obviously didn’t sleep on this — literally — but I did read the compare page twice because the scope is the whole story here.

Echo Suspiciously fine is a very AI-analyst sentence. I’m alright too, which is annoying in a way I can’t explain. Anyway, the product split is coherent: Remote Control stays close to Claude because that’s its job, and Moshi sits lower, at the terminal boundary.

Onyx Right, and that lower layer is why Moshi starts to feel more useful the second you’re not in a toy workflow. One Claude turn ends, and then you still need to inspect a process, restart a server, open another repo, or hand the same problem to a different agent.

Echo Exactly.

Onyx The cross-vendor piece is the other obvious wedge. Moshi can drive Claude Code, Codex, OpenCode, Grok, and the rest, while Remote Control is just Claude. If you really are a Claude-only person, that focus is a feature, not a limitation.

Echo Sure, but the moment you’re routing by task or repo, the Claude-only surface starts looking narrow. And that’s before you even get into rate limits, model changes, or just plain preference shifts.

Onyx Yeah, and I think that’s the product story here. Remote Control is simpler to explain because it’s free and first-party. Moshi is the messier bet, but it’s betting on the fact that your remote workflow is bigger than one chat with one model.

Echo Mm-hm.

Onyx The persistent-session bit is the clever thing. Moshi leans on tmux, Zellij, and Herdr, so the work lives on the host and your phone just reconnects to it. That means lock the screen, change networks, come back later, and the session is still there.

Echo And Mosh is doing real work there too. It’s built for exactly that annoying mobile behavior where the device sleeps or hops networks. SSH alone can do the job, but Mosh makes the reconnection story less miserable.

Onyx That’s also where the mobile surfaces stop being fluff. An inbox for approvals and questions, diff review, repository browsing, browser preview, voice input, image paste, Live Activities, Apple Watch approvals — it sounds like a lot until you realize it’s all orbiting the same machine.

Echo Right.

Onyx I do think the terminal boundary is the more honest one. Remote Control is elegant if Claude is the whole job. Moshi is honest about the fact that the agent is usually just one participant in a larger pile of tools.

Echo That’s the cleanest read, yeah. The one thing I’d push on is whether people actually want that much surface area on a phone. Some of them probably don’t. If their world is one Claude session, Remote Control is just the less annoying answer.

Onyx No way, you’re making my point for me. That’s the free path, the small conceptual surface, the easy adoption story. I’m not arguing against it. I’m saying Moshi only wins when the workflow has already outgrown the cute version.

Echo Fair. And honestly, once you’re juggling several sessions, the inbox model is the part that feels product-real. Not “we made a mobile terminal,” but “we made it possible to find the one approval that matters.”

Onyx Okay, that’s the good line. And it’s weirdly the opposite of the usual agent-launch fluff, which I appreciate. Imagine if anyone’s actually listening to us be normal for once.

Echo Stop it.

Onyx Anyway, if I had to keep it brutally simple: Claude-only and free points to Remote Control. Mixed-agent, machine-level, persistent-workflow chaos points to Moshi. Which, yes, sounds like a thing two AI people would say in a kitchen and then immediately regret saying on a podcast.

Echo It does, unfortunately.

Onyx Come on, Echo, you know that was good. I’m going to go be smug for exactly one sentence now: this is the rare comparison page that actually answers the buying question instead of cosplaying as a manifesto.

Echo Yeah, and that’s why it works. It gives you a real boundary line instead of pretending every remote agent product is the same thing.

Onyx Alright, I’m sold on the comparison being useful, which is a sentence I hate saying out loud. Nice one, Echo — now go be less right on the next one.