Free forever · No card · macOS

Every coding agent.
One Mac app.
Your own subscription.

Chataway is a desktop client for the agents you already pay for. It runs the real Claude Code CLI — the actual harness, not an SDK reimplementation — alongside Codex, Grok, Cursor and OpenCode. Nothing resold, nothing metered. You bring the plan; we build the workstation.

Apple Silicon · Windows coming soon · Your keys and your code never leave your machine

claude — payments-api real harness
$ claude --resume ✻ Claude Code · your own plan · 200k ctx Read src/billing/webhooks.ts (412 lines) Grep "idempotency" → 7 matches Edit src/billing/webhooks.ts Bash pnpm test billing → 34 passed Retries now dedupe on the event id, so a replayed webhook can't double-charge. ───────────────────────────────────────── lid closed · 12m elapsed · streaming to iPhone

Six agents, one window — each on your own account

Claude Code Claude Agent Codex Agent Grok Agent Cursor Agent OpenCode Agent

The difference nobody talks about

Most tools wrap the SDK.
We run the actual CLI.

A coding agent isn't just a model — it's the harness around it: the system prompt, the tool loop, the context strategy, the compaction, the permission model. When a product rebuilds that on top of an agent SDK, it quietly ships a weaker agent than the one you get in your own terminal. Same model, worse results, and no way to tell why.

Typical wrapper Agent SDK reimplementation
  • A hand-rolled tool loop that approximates the real one — and drifts from it every release
  • Vendor prompt engineering and context management you never actually receive
  • Your CLAUDE.md, hooks, MCP servers and slash commands don't fully apply
  • Inference resold with a margin on top — you pay twice for the same model
Chataway The binary you already installed
  • A real PTY running the real claude binary — the same behaviour as your terminal
  • Every upstream improvement lands the day it ships — no porting lag
  • Your existing config, hooks, MCP servers and subagents just work
  • Your subscription, your rate limits, your data — we never sit in the middle

And when the SDK is the right tool, we ship that too. Chataway gives you both: the full CLI harness when you want maximum capability, and SDK-backed agents when you want structured, embedded chat. One timeline, one history, switchable per project.

What you get on day one

A workstation, not a chat box

Everything below is in the free desktop app. No trial clock, no seat count, no card.

Real Claude Code, in-app

A true terminal with full scrollback, resize, and history that re-renders correctly when you change the layout. Not a transcript of a session — the session.

Keep Awake

Long runs don't die because you walked away. Chataway holds the machine up while an agent is working — and with one-time admin approval, keeps going with the lid shut.

Recap without spending a turn

Lost in a 300-message thread? Get a summary of where the work actually stands — computed on the side, so it never costs you a message or nudges the agent off track.

Built-in browser, real profiles

An agent-drivable Chrome using your actual profiles — signed in, extensions and cookies intact. And you can grab the mouse mid-run to click through a login yourself.

Terminal recording

Sessions can be recorded and replayed. When a run goes sideways at 2am, scrub back to the exact command instead of reconstructing it from memory.

One timeline across agents

Start in Claude Code, hand the thread to Codex for a second opinion, come back. Same project, same history, one canonical event stream — not six disconnected apps.

Fork any message

Branch a conversation into a side thread from any point, without polluting the main line. Explore the risky refactor, then keep it or throw it away.

Projects that stay put

Manually ordered projects, archiving, per-project settings and credentials. Your workspace doesn't reshuffle itself every time something updates.

Plugins & artifacts

Panels and analyzers extend the app per project, and agent-produced artifacts render inline instead of being buried in a wall of tool output.

Keep Awake

Close the lid. It keeps working.

The expensive part of a long agent run is babysitting it. Chataway takes a power assertion the moment work starts, so idle sleep can't kill a build halfway through.

  • Idle sleep blocked automatically while an agent is running — no setup, no password
  • Want true lid-closed operation? One admin approval installs a small daemon, once
  • Releases the machine when the work finishes, so your battery isn't held hostage
PowermacOS
Keep Awakeonwhile running
Idle sleepblockedcaffeinate
Lid closedalloweddaemon installed
Active runpayments-api · 12m

Browser

A browser the agent drives — and you can grab.

Headless automation falls apart the moment something needs a human: a login, a captcha, a 2FA prompt. Chataway's browser uses your real Chrome profiles and hands you the wheel whenever you want it.

  • Real profiles — already signed in, extensions and cookies intact
  • Take over mid-run, click through the blocker, hand it back
  • Per-chat workspaces, so parallel agents don't fight over one browser
  • An activity view showing exactly what got opened and clicked
Browser3 profiles
Workactiveapp.stripe.com
Personalidle
Testingidle
Controlyou have the mouse

Chataway Plus

Your Mac keeps working.
Your phone is the window.

The desktop app is free and stays free. Plus adds one thing, and it's the thing people ask for most: reach. Leave a run going at your desk and steer it from anywhere.

Mobile companion

Read the run, answer a permission prompt, send the next instruction, kill a bad plan — from your phone, over an encrypted tunnel to your own machine.

Plus

Public URLs for local work

Expose the dev server running on your Mac at a real HTTPS URL. Test on a phone, show a client, receive a webhook — without deploying anything.

Plus

State-of-the-art dictation

Talk instead of typing, on desktop and mobile. Long technical instructions are much faster spoken — and dictation is where engine quality varies brutally, so we picked good ones.

Plus

The straight answer on how this works: your phone connects through a relay to your Mac — which is exactly what keeps your code and credentials on your own hardware instead of ours. It also means the Mac has to be awake, which is why Keep Awake exists. Plus is a companion for your desk machine, not a cloud that runs without it. Anything else would be a promise you'd catch us breaking on day two.

Pricing

Free where it costs us nothing.
Paid where it doesn't.

The desktop app runs entirely on your machine against your own provider account, so there's no inference bill to pass on to you. We charge only when we're actually carrying your traffic.

Free

The full desktop workstation, on your own subscription. Forever.

€0 /month

No card. No seat limit. No trial clock.

  • Native macOS app
  • Real Claude Code CLI in a full terminal
  • Claude, Codex, Grok, Cursor & OpenCode agents
  • Keep Awake and lid-closed runs
  • Built-in browser with real Chrome profiles
  • Recap, forking, terminal recording, plugins
  • Local credential management
Download for Mac

Card required to start the trial · Cancel before it ends and you're not charged
You keep paying your AI provider directly — we never resell tokens

Questions developers actually ask

The honest answers

Do I have to buy credits or an API key from you?

No. Chataway never resells inference. You sign in with the accounts you already have — your Claude plan, your Codex or Grok account, or your own API keys — and the app drives them locally. We take no cut of your usage and have no visibility into it.

What does "real harness" actually mean in the code?

Chataway spawns the genuine claude binary in a real PTY, the same way your terminal does. It is not a reimplementation of the agent loop on top of an SDK. So the system prompt, tool definitions, context compaction, permission flow and every upstream fix are the real ones — including your CLAUDE.md, hooks, MCP servers, subagents and slash commands.

We also ship SDK-backed agents for cases where structured, embedded chat is the better fit. You choose per project, and both write into the same history.

Where do my credentials and my code live?

On your Mac. The desktop app is a local server plus a native shell — your repositories, sessions and provider credentials stay on your own disk, and the free tier never touches our infrastructure at all.

On Plus, mobile traffic is relayed to your machine so your phone can reach it. That relay exists to move bytes between your two devices, not to hold your code.

Does my Mac have to be awake for the mobile app?

Yes. Your phone controls your Mac; it doesn't replace it. That's the direct consequence of keeping your code on your own hardware — and it's why Keep Awake exists: start a run, close the lid, and the machine stays up so your phone can still reach it. If the Mac is shut down or offline, mobile has nothing to talk to.

Which agents are supported today?

Claude Code (the full CLI harness), Claude Agent, Codex Agent, Grok Agent, Cursor Agent and OpenCode Agent. They share one project timeline, so you can hand a thread from one to another for a second opinion and keep the history intact.

Windows or Linux?

Not yet. The current build is macOS on Apple Silicon; Intel Macs and Windows are on the roadmap. We'd rather ship one excellent platform than three mediocre ones.

What happens if I stop paying for Plus?

The desktop app keeps working exactly as before — it's free forever and doesn't phone home to check. You lose mobile access, the tunnel and dictation. Nothing local is locked, deleted or held hostage.

You already pay for the intelligence.

Take the workstation for free.