Gentler Streak Shipped Siri and On-Device AI in iOS 27: Why App Intents Are Not an Agent Bridge
A health app shipping on-device AI and Siri support reads, at a glance, like a step toward the agentic future. It is not. App Intents wire an app to Apple's own stack: Siri, Shortcuts, Spotlight. They do not wire it to the agent you actually use, and that gap is worth naming precisely.
Quick answer
Gentler Streak shipped Siri App Intents and on-device AI workout titles in version 5.13, announced in its newsroom on 14 September 2026 (evidence E9 in our weekly market pack, read 2026-10-18). Good product work. But App Intents are a system bridge, not an agent bridge: they expose an app to Siri, Shortcuts and Spotlight, inside Apple's platform. They expose nothing to Claude, ChatGPT or Cursor, and they ship no export. An agent bridge is a documented interface, such as the MCP server the agent calls, that runs outside Apple's stack.
What App Intents actually do
App Intents are Apple's framework for letting your app participate in the system. An app declares actions and entities, and then Siri can invoke them, Shortcuts can chain them, and Spotlight can surface content. If you say "log a workout" or "show my activity", Siri can hand that to the app.
That is a real integration and it is the right tool for the problem it solves. It solves app-to-system. It does not solve app-to-agent, because the reachable surface is Apple's, and the models behind Siri are Apple's choice, not yours.
Why it feels like an agent bridge but is not
Three things make the confusion understandable:
- It mentions AI. On-device AI for workout titles is inference running locally, which is great, and it is still the app answering, not your agent.
- It is on-device. Local is the privacy property we care about, so an on-device feature reads as aligned. But local inference inside a closed app and local data exposed to your agent are different things.
- It has an interface. There is a schema, a set of actions, a documented framework. An MCP server also has a schema. The difference is who can call it.
System bridge vs agent bridge
| App Intents (system bridge) | MCP server (agent bridge) | |
|---|---|---|
| Who can call it | Siri, Shortcuts, Spotlight | Claude, ChatGPT, Cursor, any MCP client |
| Who chooses the model | Apple | You |
| Runs outside Apple's stack | No | Yes |
| Gives you the raw data | No, it runs app actions | Yes, read-only metric tools |
| Portable to another OS | No | Yes |
The column that matters is the fourth. An agent bridge hands your own data to your own agent. A system bridge runs the app's own actions for you. One is access, the other is convenience.
How we sourced this. The version, the 14 September 2026 date, and the Siri and on-device AI features are from the app's own newsroom post (E9), read on 2026-10-18. Feature lists change with each release, so check the current release notes.
Why the distinction costs you
If you pick a health app for its AI features and later want an agent to answer questions about your data, App Intents do not get you there. There is no export and no documented interface an outside model can call, so the answer is a screenshot, a manual copy, or nothing. You end up re-entering your own data into a chat window, which is the exact problem an agent bridge exists to remove.
What a real agent bridge looks like
It is boringly specific: a documented interface, a named tool list, and a data path you can inspect. MetricBridge ships a zero-dependency stdio MCP server that reads your local Apple Health JSON export and exposes 16 read-only tools. No App Intent, no Apple-only surface.
{
"mcpServers": {
"health-export": {
"command": "node",
"args": ["/Users/you/.health-export-mcp/server.mjs"],
"env": {
"HEALTH_DATA_DIR": "/Users/you/Library/Mobile Documents/iCloud~ai~healthexport~app/Documents"
}
}
}
}Then ask your agent to call get_mcp_status, and ask it how your HRV trended this week. Same data, same local machine, but the model answering is the one you chose, not the one bundled with your phone. Everything it reports is a deviation from your own baseline, not a diagnosis.
The test to apply to any health app in 2026
Ask one question: can an AI agent I did not choose read my data through a documented interface? If the answer is no, the app has a system bridge at best. If the answer is yes, and the interface is open source and reads a local file, you have an agent bridge. That single question separates the two categories faster than any feature list.
FAQ
What are App Intents in iOS?
App Intents are Apple's framework for letting an app expose actions and content to the system, so Siri, Shortcuts and Spotlight can invoke them. They connect the app to Apple's platform. They do not expose the app to third-party AI agents.
Does Siri support mean my health app works with Claude or ChatGPT?
No. Siri support is a system bridge inside Apple's stack. An agent bridge is a separate, documented interface, such as an MCP server, that a third-party client like Claude, ChatGPT or Cursor can call. The two are unrelated.
Gentler Streak has on-device AI. Is that private?
On-device inference keeps that computation local, which is the right instinct. It is still the app's own model answering inside the app. It does not give your own agent access to your raw data, which is a separate question.
How do I get my Apple Health data into an agent then?
Use an app that exports your data to a local file and ships a documented interface, such as an MCP server, that reads it. MetricBridge exports 190 metrics as local JSON and its open-source stdio server exposes 16 read-only tools an agent can call.
Doing this yourself? MetricBridge handles the export step: all 190 metrics as local JSON, with the open-source MCP server (GitHub).
An agent bridge, not a system bridge
MetricBridge is a $24.99 one-time unlock: 190 metrics as local JSON, the open-source MCP server your agent actually calls, and the on-device Ask and Weekly Brief.