If you run coding agents on more than one machine, you already know the most annoying failure mode in the whole setup: the agent stops, asks a question, and waits. You are not at that keyboard. The work sits idle until you walk back, unlock the machine, and press Enter. The promise of Inyon is simple — to let you answer coding agents from phone, wherever you are, without changing how the agents run.
Inyon is a free, open-source tool from solo maker Abhash Chakraborty that turns coding agents already running in tmux on your machines into chats on your phone or in a browser. When an agent asks for permission, shows a menu, or asks a question, that prompt becomes a button. You tap, and the answer is typed into the same terminal pane — so when you sit back down, nothing has changed except that the agent is moving again. This review covers what Inyon does, what it costs, where it falls short, and whether it belongs in your workflow.
What Inyon actually does
Inyon does not run your coding agents. That distinction matters, because most tools in this space want to own the whole session: they host the agent, they hold your account, they become the place your code lives. Inyon takes the opposite approach. Your agents keep running exactly where they already run — in tmux sessions on your laptop or servers, under whatever subscriptions or API keys those machines already have. Inyon only watches those sessions and gives you a way to answer them remotely.
The maker built it out of his own frustration. He runs coding agents on a laptop and two servers, each signed in to a different account, and kept finding an agent stalled on a question like “can I run the tests?” while he was at a different machine. The agents’ own remote features follow one account at a time, which did not fit a multi-machine, multi-account setup. Inyon was the fix: one phone view across all of them.
Each tmux session appears as a chat. You see what the agent is doing, and when it needs you, the prompt arrives as a tappable button instead of a terminal line you have to read and retype. That is the whole interaction model, and its restraint is the point. There is no new agent to learn, no session to migrate, and no dashboard that duplicates your terminal.
How to answer coding agents from phone with Inyon
Setup is one line per computer, using the installer from the project’s site. There is nothing to host: no server to provision, no database to babysit, no reverse proxy to configure. Once installed on a machine, you pair the Android app with that computer by scanning a QR code. The app then reaches your machines over an end-to-end encrypted tunnel, or over your own VPN if you prefer that route.
- Install the Inyon agent on each machine where coding agents run, one line per computer.
- Pair the Android app with each machine by QR code — no account, no sign-in anywhere.
- Open a session chat when an agent stalls, and tap the button for the permission, menu choice, or answer it needs.
- Your tap is typed into the same tmux pane, so the session continues exactly as if you had answered at the keyboard.
- Use the browser view when your phone is not handy; the same session chats are available there.
That “typed into the same pane” detail is worth pausing on. Some remote tools inject answers through a side channel, which can leave the terminal state subtly different from what you expect when you return. Inyon’s approach keeps the terminal as the single source of truth. What you see on the phone is the session, and what you tap goes into the session — not into a copy of it.
Which agents it supports — and the beta caveat
Inyon supports six coding agents: Claude Code, Codex, Gemini CLI, OpenCode, Copilot CLI, and Antigravity. That is a strong list for a solo-built tool, and it covers the agents most professional teams actually run day to day. If your whole fleet is Claude Code and Codex, you are squarely in the supported middle of the road.
The honest caveat, stated by the maker himself, is that three of the six agent readers are still in beta. The Product Hunt page does not name which three, so treat any agent outside the most popular two as “try it on a non-critical session first.” Beta readers in this kind of tool can misread a prompt, miss a menu option, or present a question in a form that loses context. For answering “run the tests?” that is low stakes. For approving a destructive command, it is not — which brings us to the safety question every team should ask before adopting any remote-approval tool.
Remote approval is only as safe as the approval itself. Tapping “yes” from a phone notification is faster than walking to a laptop, and speed is exactly what makes it riskier: you are more likely to approve without reading the full command. Our rule for any tool like this is to use it for low-risk answers — test runs, file reads, menu choices — and to keep destructive or irreversible actions for a proper keyboard session where you can read the whole diff. Inyon does not change that rule; it just makes following it more convenient.
Pricing: free, open source, and no account
Inyon is free and open source under the MIT licence, listed at 0 USD on Product Hunt. There are no tiers to compare, no per-seat pricing, and no subscription to evaluate — which is itself a statement about the tool’s philosophy. Each machine keeps its own subscription or API key, so Inyon never becomes a middleman for your agent spending, and it never asks you to sign in anywhere.
Two design choices back up that privacy posture. Releases are signed, and the npm package is published from CI with provenance, so the code you install is verifiably the code the project shipped. And because there is nothing to host, there is no Inyon cloud holding a copy of your sessions, prompts, or code. Your data travels between your phone and your machines over an encrypted tunnel or your VPN — not through a vendor’s servers.
For a business evaluating tools, that architecture removes an entire category of procurement questions. There is no data-processing agreement to negotiate, no vendor access to your codebase to assess, and no subscription cost to justify. The costs that remain are the ones you already pay: your agent subscriptions, your machines, and the time to try it. That is a genuinely different proposition from most agent tooling, and it is Inyon’s strongest selling point.
Who it is for — and who should skip it
Inyon fits best if you already run coding agents in tmux on one or more machines and you regularly step away while they work. Solo developers with a laptop and a home server, agency developers juggling client projects across machines, and anyone running long agent sessions overnight will feel the benefit immediately. The multi-account angle matters too: if your machines are signed in to different accounts, the vendors’ own remote features may not cover you, and Inyon’s account-free design fills that exact gap.
Skip it, for now, if you are on iOS. There is no iOS app yet, and that is a hard blocker, not a minor gap — the whole product is the phone experience. Also skip it if you do not use tmux, or if your agents run inside an IDE rather than terminal sessions: Inyon reads agents already running in tmux, so a non-tmux workflow has nothing for it to attach to. And if you need guaranteed support for one of the beta readers on production work, wait until the maker marks it stable.
Teams should also weigh the solo-maker factor honestly. Inyon is built and maintained by one person, which means fast decisions and a clear vision — but also a bus factor of one and a support channel that is essentially the project’s own issue tracker. For a free tool that touches your development sessions, that trade-off is reasonable. For a tool you plan to standardise across a whole engineering team, pilot it with a few developers first and watch how quickly issues get answered.
How it compares to the alternatives
The most direct alternatives are the agents’ own remote features — the companion apps and web views that vendors ship for their own tools. Those are usually polished, but they follow one account at a time and they only cover that vendor’s agent. If you run Claude Code on one machine and Codex on another, under different accounts, no single vendor app gives you one view. Inyon’s whole pitch is that unified, vendor-neutral view.
A second family of tools runs the agent for you in a hosted session — you open a browser terminal, pick a repo, and the agent works in the cloud. That model is convenient, but it moves your code and your agent spend into someone else’s infrastructure, and it changes how the session works. Inyon deliberately does neither: your agents stay where they are, and the phone is only a window into them.
A third, quirkier comparison is the wave of agent-monitoring gadgets — browser extensions that surface agent status in a tab, for example. Those solve the “which agent needs me” problem on your desktop. Inyon solves the same problem when you are not at your desk at all. They are complements, not competitors: one watches from your browser, the other answers from your pocket.
From our experience: where tools like this actually pay off
Having shipped software with AI-assisted workflows for years at MTD Technologies, we have learned that the bottleneck in agent coding is rarely the model’s intelligence. It is the handoff: the agent does ninety minutes of solid work, then waits forty minutes for a human to confirm something trivial. Multiply that by several agents across several machines, and the idle time quietly eats the productivity gain you bought the agents for in the first place.
Tools that attack the handoff, rather than adding another agent, tend to age better than the flashier launches. They have fewer moving parts, fewer ways to break, and a benefit you can measure in a week: how many stalled sessions got unblocked without a trip to the desk. That is the right way to evaluate Inyon. Run it for a week on your real sessions, count the unblocks, and note any prompt it misread. If the count is high and the misreads are zero, it has earned its place. If you find yourself re-checking every tap at the keyboard anyway, it has not saved you anything.
One practical gotcha to plan for: notification discipline. An agent that can reach your phone can also interrupt your evening with every question it has, including the ones that could have waited. Decide upfront which sessions get phone access, and keep your most experimental agents off it until you trust their judgement. The tool gives you reach; you still have to supply the boundaries. If you want help wiring agent workflows like this into a real delivery pipeline, our AI integration and automation services cover exactly that kind of setup — and our AI tools directory tracks the launches worth watching.
Frequently asked questions
Is Inyon really free?
Yes. Inyon is free and open source under the MIT licence, listed at 0 USD on Product Hunt. You still pay for your own agent subscriptions or API keys on each machine, but Inyon itself charges nothing and requires no account.
Does Inyon work on iPhone?
Not yet. There is no iOS app at launch — the phone experience is Android only, plus a browser view that works on any device. iPhone users will need to wait for an iOS release or use the browser view.
Which coding agents does Inyon support?
Six: Claude Code, Codex, Gemini CLI, OpenCode, Copilot CLI, and Antigravity. Three of the six agent readers are still in beta, so test your specific agent on a non-critical session before relying on it for important work.
Does Inyon host my code or sessions?
No. There is nothing to host and no account to sign in to. Your agents keep running in tmux on your own machines, and the phone app reaches them over an end-to-end encrypted tunnel or your VPN.
Do I need tmux to use Inyon?
Yes. Inyon reads agents already running in tmux, so your coding agents need to be running in tmux sessions on the machines you pair. If your agents run inside an IDE instead, Inyon has nothing to attach to.
The verdict is straightforward: if you run coding agents in tmux across more than one machine and you carry an Android phone, Inyon removes the single most annoying friction in that setup for free, without asking for your data or your account. The beta readers and the missing iOS app are real limits, but they are the kind that shrink with time, not structural flaws. Try it on a real session this week — the first time an agent unblocks itself while you are away from your desk, you will know whether it stays. And if you want a hand setting up agent workflows that fit your team rather than fighting it, talk to us — that is the work we do every day.