Skip to main content

Work on your Mac. Keep going from your phone.

Let Codex or Pi work with files on your Mac while you follow the conversation in NanthAI. Review the reply, ask for a change or bring in a cloud model from your phone or browser.

Start in the cloud. Bring in your Mac when you need it.

OpenRouter

Choose cloud models for writing, research and review directly in NanthAI. Get started from your phone or browser.

Codex

Use your existing Codex account to work on code and files on your Mac. Send the task from NanthAI and follow its progress wherever you are.

Pi

Use Pi with your configured providers and subscriptions, including OpenRouter, or with local models through Ollama. Bring its work into the same conversation as your other Personas.

Bring the next model up to speed.

Keep published replies, files and useful memories with the conversation. Ask a cloud Persona to review a local result, or continue with another local Persona. Each participant keeps its own private working history.

Keep a hand on the work.

Answer a runtime question or permission request from NanthAI. Stop a task when you need to, and see the reported model usage alongside the work.

Let the next task start on schedule.

Schedule a recurring brief or a sequence of tasks. Choose the runtime for the work and find the results in NanthAI when it finishes. Keep your Mac online for local runs.

Connect once. Keep working from your phone.

  1. Get the Mac app

    Install NanthAI Runtime and choose Codex, Pi or both.

  2. Connect your account

    Link your Mac to NanthAI once. Your conversations can now reach it.

  3. Send your first task

    Choose a Local Persona, tell it what to do, and follow the reply from your phone.

Set up your Mac

Your Mac needs to be on and connected for tasks sent from your phone. Cloud models need internet; Pi can also use models already installed through Ollama. Available tools depend on the runtime and model you choose.

Give your AI the context that matters.

NanthAI brings relevant conversation history and memories into each turn. In our testing, selected Pi scenarios used up to 64% fewer input tokens and had 29% lower estimated model cost.

Fewer input tokens

Up to64%

Pi alone · 100%

Pi with NanthAI · 36%

Lower estimated model cost

Up to29%

Pi alone · 100%

Pi with NanthAI · 71%

Observed with Pi and GPT-6.1 Sol at low reasoning, compared with Pi alone. These highlights come from a context-heavy follow-up, not an average across all tasks. Results vary by task and model. Cost is an API-equivalent estimate, not a reduction in subscription charges.