Ask Claude “How did I sleep this week, and did the late dinners make a difference?” and it answers from your real nights, not from a guess, because it can read them from Voke. It reads only what you approve, and changes only what you allow.
You can set this up with the assistant you already use.
Which assistants work
Voke connects to any AI app that supports MCP, the open standard assistants use to reach other services. In Voke, open the AI tab, tap the settings button, and you’ll find these under AI apps, each with its own steps:
- Claude, set up on the web or in Claude Desktop. Step-by-step guide.
- ChatGPT, through developer mode on the web. Step-by-step guide.
- Gemini, Grok, Perplexity and Mistral Vibe, each through its own connectors settings.
- Grok Bot, which you ask in chat to add Voke. How it works.
- OpenClaw and Hermes Agent, the AI agents you run on your own computer. Step-by-step guide.
- Developer tools, with install links for VS Code and Cursor and a one-line command for Claude Code.
- Any other MCP app. If it lets you add a remote MCP server, paste Voke’s address:
https://mcp.voke.co/mcp.
Voke reads from Apple Health and from Google Health, which is where your Fitbit or Pixel Watch data lives. So whichever assistant you pick, it can answer from your Apple Watch, your iPhone, your Fitbit, or all of them at once.
What you can ask
Your assistant gets a handful of tools from Voke. You don’t need to know their names. Just ask the way you’d ask a friend who can see your data:
- “Give me a plain summary of today: sleep, activity, anything that stands out.”
- “How did my sleep this week compare with the week before?”
- “What was my heart rate doing during Tuesday’s run?”
- “Is my resting heart rate creeping up this month?”
- “Which nights did I get the least deep sleep, and what did those days have in common?”
- “Is my watch still syncing?”
Each request covers up to 31 days, or 14 days for minute-by-minute data. Your assistant can ask again for an earlier stretch, so longer questions still work. They just take a few more steps.
If you keep a journal in Voke, your assistant can read your tags and notes too, so it can line up late meals or travel days with how you slept. The journal starts switched off when you connect an app, so it’s only shared if you turn it on.
How connecting works
Every assistant follows the same three steps:
- Add Voke’s address in the assistant’s connector settings, or with one command for agents.
- Approve on your phone. The assistant opens a Voke page with a QR code and a 6-letter code. Scan the QR code with your iPhone camera, or tap Enter a code in Voke and type it in.
- Choose what to share. Voke shows “Connect Claude?” (or whichever app it is) with a switch for each kind of data. Turn off anything you’d rather keep to yourself, then tap Allow.
You never type your Voke password into the assistant, and it never sees it. Voke uses OAuth, the standard way for one app to let another in without sharing a password, so the assistant only gets a key to the data you approved. The page lapses after 10 minutes, and you can start again any time.
What your assistant can and can’t see
- Read, or read and change. For each kind of data you choose whether it can only read it or can also add and change entries.
- Only the kinds of data you switched on. Turn off heart, and it can’t read your heart data at all.
- Your own data only. It reads what you’ve connected to Voke, not what people in your circles share with you.
- Every read is logged: which app, which kinds of data, and whose.
Where your data goes next
Once your assistant reads something from Voke, a copy of it sits in that chat, and the assistant company’s own privacy terms apply from then on. Voke can’t reach into their app to delete it. If you want it gone, delete the chat there.
Each company also has its own setting for whether your chats help train its models. It’s worth a look before you start:
- Claude doesn’t train on what it reads through connectors like Voke, unless you copy it into the chat yourself. The rest of your chat follows the model improvement setting in Claude’s privacy settings (Anthropic).
- ChatGPT may train on what it reads from connected apps on Free, Go, Plus and Pro if “Improve the model for everyone” is on. You’ll find it under Settings, then Data controls (OpenAI).
- Gemini needs Keep Activity turned on before you can add a custom app like Voke (Google). With Keep Activity on, Google may use your chats to improve its models, and people may review some of them (Google).
- Grok Bot follows your Cursor account’s privacy settings (xAI).
- OpenClaw and Hermes Agent run on your own computer and send your chats to whichever AI model you set them up with, so that model provider’s terms apply.
Voke never sends your data to an assistant on its own. The assistant only gets what it asks for, inside what you approved. The rest of how Voke handles your data is in how Voke keeps your health data private.
A few habits that help
- Share what the question needs. If you only want to talk about sleep, connect with just sleep switched on. You can connect again later with more.
- Treat answers as a starting point. An assistant can spot patterns and explain them in plain words. It isn’t a doctor, and it can be wrong. If something worries you, take it to one.
- Check your AI apps now and then. Anything connected shows “Connected” in the list, so a forgotten connection is easy to spot.
How to disconnect
Right after an assistant connects, Voke sends a notification. Press and hold it for a Disconnect button. That’s the fastest way out if you didn’t expect it.
Any time after that:
- One app. In AI apps, tap Disconnect next to it. Its access ends on its next request, and the others keep working.
- Everything at once. In your account, turn off “Use AI features”. That disconnects every app, turns off Ask Voke, deletes your chats and keeps your data out of other people’s AI too.
A connection lasts 90 days. After that, your assistant asks you to approve it again on your phone. Removing Voke in the assistant’s own settings only tidies up its side. Disconnecting in Voke is what actually ends the access.
Connecting your own AI works in the Voke beta today. Ask for an invite and try it with the assistant you already use.