Supercomputer power.
Made for everyone.

Speak to your Spark and hear its replies. Continue a conversation or give voice instructions to an existing Oh My Pi or OpenCode session on your Spark. CarPlay is implemented in the development code, not released yet. Apple's CarPlay entitlement approval and real-car testing are still pending. Requires a reachable Spark and configured voice services. Coding sessions need Engineering Mode, project access and a session already running on the Spark. Set up and review your work on iPhone before driving; the car interface uses voice, not a code editor.

We're adding watchOS support to bring SparkPilot to Apple Watch. Watch features and release timing are not confirmed yet; watchOS is not part of the current beta.

Newer development work includes Spark-hosted coding workspaces with Engineering Mode, reconnectable terminal sessions, searchable chat history, plain-style replies, share-sheet imports, Spotlight and follow-up chats from automation results. These join automations, guest sharing, calendar context, charts and optional web answers. These are findings from development branches, not a confirmed TestFlight release checklist; check the TestFlight release notes for what is included.

The newer development workspace supports this through Engineering Mode. Your project files and coding-agent process live on the Spark while SparkPilot provides the interface on your Apple device. Oh My Pi, OpenCode, Pi and Codex have routes to a supported Spark model. The separate Claude Code option uses your Anthropic account and cloud models, so it is not local inference. A compatible runtime, project permissions and tool setup are required; this is not yet a verified TestFlight feature.

No. Regular chat history is saved on the Apple device where you used it, not synced or included in device backups. Project files in a development workspace are different: they live on the Spark and can be reopened from an authorised device with access to that project.

No. They are interactive explanations with clearly labelled fictional examples, not app screenshots, live AI or benchmarks. They do not record audio, connect to a Spark or save your choices. The separate app preview uses simulator screenshots, and the existing text-chat video is labelled as a real simulator recording.

Yes, through guest sharing in the newer development build. Invite a friend, family member or colleague and choose which models they can use. Their guest view identifies the Spark as shared with them and lists only those models, not your full library. You confirm their device and control their schedule, daily allowance and expiry; they cannot administer the Spark. Both people need compatible SparkPilot builds and a connection to the Spark, on the same network or remotely through a separately configured Tailscale connection. Sharing does not transfer ownership.

Yes. Read the public development updates on this page or follow the RSS feed at https://sparkpilot.dev/updates.xml. You can join through our public TestFlight link without submitting an email address on this website.

SparkPilot is an independent, free app for managing an NVIDIA DGX Spark from an iPhone, iPad, or Mac. It brings guided setup, model management, text and voice chat, optional Tailscale access, cluster setup, Apple Foundation Model Intents, Siri and Shortcuts, hardware controls, and Home Screen widgets into a native Apple app. On Mac, launch Hermes Agent from SparkPilot or connect other coding tools through an authenticated OpenAI-compatible endpoint to the model on your Spark. SparkPilot is in beta, with a public TestFlight join link on this website.

Yes. No subscription, paid tier, or ads, ever. You need access to a DGX Spark, either your own or one shared with you through a compatible build, but the app won't cost you anything.

Your AI runs on your Spark, and no telemetry leaves the app or your Spark. We don't receive it or send it to third-party analytics. You can use SparkPilot over your local network with remote access switched off. Model downloads, software updates and explicitly enabled web answers or web-reading automations use the internet. Apple separately collects TestFlight beta crash and usage data and shares it with developers.

Join SparkPilot through TestFlight. Open the link on your Apple device and follow Apple's instructions. No email request needed. TestFlight shows the available builds and device requirements. Join at https://testflight.apple.com/join/qpW2Swh2. Beta places and compatible builds are subject to availability.

Join SparkPilot through TestFlight. Open the link on your Apple device and follow Apple's instructions. Beta places and compatible builds are subject to availability. No email request needed. TestFlight shows the available builds and device requirements. Beta places are limited.

We'll only use your email address to send your SparkPilot TestFlight invitation. We'll never send you junk or marketing emails, sell your address, or share it for any other purpose. Cloudflare and our email provider process your request so it reaches us, and Apple processes your address to deliver the TestFlight invitation. Your request is stored on Cloudflare for up to 30 days to prevent duplicate submissions, and a copy stays in our inbox until deleted. Email hello@sparkpilot.dev to ask us to delete your request.

TestFlight is Apple's free app for trying beta versions before release. Install TestFlight on the Apple device you want to use, open our public join link, and follow Apple's instructions. Beta builds may contain bugs. Apple collects beta crash and usage data through TestFlight and shares it with developers, separately from SparkPilot's local Spark telemetry.

SparkPilot is coming to iOS, iPadOS, and macOS. You'll need an NVIDIA DGX Spark with Spark Runtime installed. We're currently building for iOS 26, iPadOS 26, and macOS 26; these requirements may change before release.

You shouldn't have to be. That's why we're building SparkPilot. The app walks you through setup, then gives you a place to chat and manage your Spark without learning terminal commands.

We're aiming for 10 minutes from unboxing to your first chat. We haven't benchmarked that yet. Updates, your connection speed, and the size of the model download can all add time.

Yes, the Mac development build includes an authenticated OpenAI-compatible endpoint for supported coding tools and agent harnesses. Tools such as Oh My Pi support configuring an OpenAI-compatible provider. It forwards requests to the model running on your paired Spark. The endpoint is local to your Mac, and SparkPilot must stay open while a tool uses it. Setup and compatibility depend on the tool and model; this is not a promise of one-click setup or support for every OpenAI API feature.

In the Mac development build, install Hermes Agent, choose your paired Spark and a compatible model, then launch Hermes from SparkPilot. The app starts Hermes and configures its authenticated local API connection, routing inference requests to your Spark. Keep SparkPilot open while Hermes runs.

Yes. Apple Foundation Model Intents are available in the current SparkPilot development build, letting you use AI from your Spark in Apple workflows. Ask your Spark to draft an email, summarise a document, or rewrite text from your iPhone or iPad, then use the result in a Shortcut with supported app actions. You choose the workflow and grant any permissions it needs; SparkPilot does not take unrestricted control of your device or send emails on its own. SparkPilot also includes App Intents for Siri and Shortcuts to check status and load or unload a model.

Yes. The development build can use an existing Tailscale connection to your paired Spark. Install Tailscale, sign in and approve both your Spark and Apple device on the same network, then check and enable the private route in SparkPilot. The app does not create your Tailscale account or enrol devices for you. Remote access is optional and off unless you enable it.

Yes. The development build has voice conversations, dictation and read-aloud replies. English speech recognition runs on your Apple device and needs supported on-device recognition plus microphone and speech permissions. Your model runs on the Spark, and Kokoro on the Spark creates the spoken audio. Install and enable Kokoro through the app for spoken replies.

The development build can manage several paired Sparks, review existing cluster groups and guide two-Spark ConnectX setup with compatible runtimes and a physical link. Tailscale provides remote access; it does not replace the ConnectX link between cluster members. Running a single model across both Sparks is still disabled pending hardware qualification, so pooled inference is not available in the current build.

The feature directory describes the current development build, including chat and voice, models, Tailscale remote access, cluster setup, Mac agent launchers, Apple workflows, widgets and device controls. Some features require extra software, permissions or compatible models and hardware. Check the TestFlight release notes for what is included; features and requirements may change, and beta builds may contain bugs.

The Home Screen widgets show memory use, connection, and model status for a Spark or group of Sparks. You can choose an overview or memory widget. Each widget shows the latest reading saved by the app, with a timestamp. Widgets are shown from the development build.

No. We've made real simulator screenshots clickable so you can look around. The demo doesn't connect to hardware or run a model.

No. We're an independent project for DGX Spark owners, with no NVIDIA affiliation, endorsement, or sponsorship.