Index 01

Index 01 is a ring with one button and a microphone. Press the button to record a voice note. The ring sends the recording to the Pebble mobile app over Bluetooth, and the app transcribes it, acts on it, and keeps it in the Index feed. Developers can receive each recording at their own HTTP endpoint with a webhook, or give the assistant new tools by connecting an MCP server.

How the Ring and the App Split the Work

The ring firmware is a small fixed loop: it detects button presses, records audio while the button is held, holds recordings until the phone collects them, and transfers them over Bluetooth. It does not transcribe or interpret anything. All of the logic lives in the Pebble mobile app, which is open source at github.com/coredevices/mobileapp. The ring code is in the experimental module of that repository. This section documents the app side only; there is no ring firmware SDK.

Gestures

The button recognises five gestures. Each one is routed from the Ring Button screen in the Index 01 settings of the Pebble mobile app. The three music gestures can be configured on Android only; on iOS they are shown disabled.

The two hold gestures record a voice note. Index agent transcribes the note and takes actions such as creating a note, a reminder, a timer, or a calendar event. Web search answers a question and puts the answer in the feed. MCP sandbox runs the recording through a sandbox group of MCP servers that you choose. Webhook only sends the recording to your endpoint and does nothing else. A recording gesture routed to Index agent, Web search or MCP sandbox can also send a copy to its webhook with the Also send to webhook switch; the switch is not shown for Webhook only or Nothing.

Local and Cloud Processing

Where a recording is processed is a per-phone setting. The two steps are configured separately in the Index 01 settings.

Speech Recognition turns audio into text. Cloud only sends the audio to Core Devices' transcription service. Cloud, with local fallback uses the cloud and falls back to an on-device model (a 400 MB download) when the phone is offline. Local only always uses the on-device model. On iOS, iOS Speech Recognition uses Apple's on-device recogniser.

Agent Model runs the assistant that reads the transcript and takes actions. Cloud LLM runs in the cloud. Local LLM runs on the phone; it is experimental, does not support every built-in action, and cannot call remote MCP servers. Cloud LLM (with Local fallback) uses the Local LLM only when the cloud cannot be reached.

Webhooks do not depend on either setting: the audio is sent as recorded, and the transcript is whatever the selected speech recognition produced.

Building on Index 01

There are two extension points.

• Webhooks send each recording as a multipart HTTP POST to a URL you configure, with the audio, the transcript, or both, and an optional HMAC-SHA256 signature. Use this to feed recordings into your own service, an automation tool, or a note-taking app.
• MCP servers add tools to the assistant. The app connects to any MCP server over HTTP, lists its tools, and lets the Cloud LLM call them when it processes a recording.

Where the Code Lives

Everything in this section is implemented in the Pebble mobile app at github.com/coredevices/mobileapp: