Side by side
| Cadence | One (getone.one) | |
|---|---|---|
| Mac | macOS 13 or later, Apple Silicon only | macOS 13 or later (14 recommended), Apple Silicon or Intel |
| Price | Free | Free: 25 AI requests a day. Pro: $12.99 a month, or $119.88 a year. Lifetime: $229, then $299, limited to 260 copies |
| Account | None | Sign in with Apple or Google |
| Where AI requests go | Talk routing, titles, recaps and the Director run on Claude through your own Claude Code login | Through One's server to AI providers: your ChatGPT subscription first, OpenAI directly if you add your own key, otherwise One's AI. Voice, titles and recaps use One's AI |
| Speech to text | On your Mac (NVIDIA Parakeet). Audio is never uploaded | On the Mac, OpenAI, Groq, AssemblyAI, Google or Apple, depending on device and engine. From the phone, through One's servers to OpenAI |
| Voice routing | Double-tap ⌥ (Talk) | Double-tap Option |
| Dictation into any app | Yes: hold ⌥ | Not listed on its site |
| Inbox | 1 to 9, J/K, E, two-second undo | 1 to 3, J/K, E, two-second undo |
| Claude Code | Hosted natively (real Allow/Deny, cost, worktree diffs), or in herdr | In herdr |
| Other agents | Codex, OpenCode, Pi, Omp, Amp, Grok, Cursor, Copilot, Gemini and Aider, through herdr | Codex, OpenCode, Pi, Omp, Amp and Grok, through herdr |
| Phone | iPhone app, private TestFlight beta (iOS 17 or later), with a waitlist | iPhone and iPad app, free TestFlight beta (iOS 17 or later). Browser remote at getone.one/remote for other phones |
| Linux servers | Over SSH from your Mac through herdr (since 0.5.7). The server does not connect to Cadence | Up to 8, through a connector that connects out, end-to-end encrypted. Agents stay reachable when the Mac is off |
| Source code | Closed | Closed |
What they share
If you have used one, the other will feel familiar. Both apps:
- show a dot per agent on the edge of the screen: blue working, amber needs you, green done;
- collect questions and finished work in one Inbox, answered with number keys, with J/K to move and a two-second undo;
- send a spoken or typed request to the right agent, or start a new one, when you double-tap ⌥;
- attach part of your screen to a request;
- run other CLIs in herdr, the free, open-source terminal runtime for coding agents;
- have an iPhone app in TestFlight and a way to see agents on Linux servers.
Your coding agents run on your own subscriptions in both. The differences are in everything around them.
Where they differ
Price and account
Cadence is free, with no account and no subscription. One needs a sign-in with Apple or Google. Its free plan includes 25 requests a day to One's AI, which handles voice, routing, titles and recaps. Pro, at $12.99 a month, lists unlimited usage, and a limited Lifetime license costs $229, rising to $299.
Where your words and screenshots go
In One, each request sends your words, a screenshot and your agents' recent terminal output through One's server to AI providers. Its privacy policy names OpenAI, Anthropic and OpenRouter, and several speech-to-text services. One says it keeps no server-side archive of audio, screenshots or terminal text.
In Cadence, speech recognition runs on your Mac and audio is never uploaded. Talk routing, titles, recaps and the Director call Claude through the claude CLI you are already signed in to, so the request goes from your Mac to Anthropic under your own login. Cadence sends anonymous usage counts you can turn off; One's privacy policy says its app analytics go to PostHog with no opt-out in the app.
The flip side: Cadence's AI features need Claude Code installed and signed in. If you only use Codex and pay for ChatGPT, One can run its routing on that subscription instead.
Claude Code and the other CLIs
Cadence can host Claude Code itself. In Cadence chat mode, permission requests are real Allow/Deny cards, you see the cost, and each agent gets its own git worktree with Diff and History tabs. One runs Claude Code in herdr like any other CLI. Cadence also supports Cursor, Copilot, Gemini and Aider, which One's site does not list.
Dictation and the Director
Cadence's same key does a second job: hold ⌥ to dictate into any app, transcribed on your Mac. It also has the Director, a Claude session that can see every agent, answer "is anything stuck?", and start, message or stop agents when you ask.
Intel Macs, phones and servers
Here One does more. It runs on Intel Macs; Cadence is Apple Silicon only. Its phone app also runs on iPad, and any phone can use its browser remote. Its server connector keeps agents on up to eight Linux servers reachable when your Mac is off. Cadence reaches servers over SSH from your Mac, so its server agents reach your phone through your Mac.
Which one to pick
One is the better pick if
- you use an Intel Mac;
- you want to check agents from an Android phone, an iPad or a browser;
- your agents run on servers and you want them reachable while your Mac sleeps;
- you use Codex and ChatGPT, not Claude Code, and want the routing to run on that subscription.
Cadence is the better pick if
- you want no account, no subscription and no daily request limit;
- you want your voice transcribed on your Mac and your agents' output kept off third-party servers;
- you use Claude Code and want real Allow/Deny cards, cost and worktree diffs;
- you run Cursor, Copilot, Gemini or Aider next to Claude Code and Codex;
- you also want dictation into every app.
Trying Cadence next to One
Both apps use a double-tap of ⌥, so quit One while you try Cadence, or change Cadence's key in Settings → Hotkey → Push-to-Talk Key (for example to ⌃ or ⌘⌥). Then:
- Download Cadence and open it. There is nothing to sign in to.
- Make sure the
claudeCLI is installed and signed in; Cadence uses that login. - Open ⋯ → New agent… on the toolbar, pick a CLI and a folder, and start an agent. If herdr is missing, the view shows the install command.
Common questions
Is Cadence a free alternative to getone.one?
Yes. Cadence is free, with no account, no subscription and no daily request limit. It has the same core pieces as One: a status toolbar, a numbered Inbox and voice routing on a double-tap of Option. It runs on Apple Silicon Macs only.
Is One (getone.one) the same as withone.ai?
No. withone.ai is an unrelated product. This page compares Cadence with One, the Mac app for coding agents at getone.one.
Does One work on Intel Macs?
Yes. One's site lists macOS 13 or later on Apple Silicon or Intel. Cadence needs Apple Silicon and has no Intel build.
Do Cadence and One send my code to their servers?
One sends each request, including a screenshot and your agents' recent terminal output, through its server to AI providers. Cadence has no AI server of its own: its routing, titles, recaps and Director use Claude through your own Claude Code login. In both apps, your coding agents talk to their own providers under your own accounts.