Use Voice Input

Set up Voice Input

1. Open Settings > Voice Input, or tap the microphone in the Composer before Voice Input is set up.
2. In Service, choose OpenAI (gpt-transcribe), DeepInfra or SiliconFlow (China) (Qwen/Qwen3-ASR-1.7B), or Custom.
3. For Custom, fill in:
   • Base URL: an https:// URL compatible with OpenAI's audio/transcriptions endpoint.
   • Model: the model name the service expects.
   • Optional Advanced Request Fields: language hints and other form fields your service documents.
4. Enter your API Key and tap Save.
5. On the first save for a service, Heeler asks Send Recordings to …?. Tap Send Recordings to confirm.

The first time you record, iOS asks for microphone permission.

Your key stays in this device's Keychain. Tap Remove Key in Settings > Voice Input to delete it.

Record a message

Tap the microphone beside Send to speak, then tap it again to stop. The text lands at the cursor in Message Agent, so you can edit it before sending.

• A recording bar shows cancel (✕), a volume waveform, and elapsed time.
• The microphone turns red while recording, and Send and Stop hide to avoid stray taps.
• While transcribing, the microphone shows a spinner; tap ✕ to cancel.
• Each recording lasts up to 2 minutes, then stops and transcribes automatically.
• Backgrounding Heeler or leaving the Composer ends the recording and transcribes it into that draft. Entering Direct Input, leaving the Agent, or a Blocked Card replacing the Composer in Chat cancels it.

Languages

Languages start at Automatic, regardless of the device's preferred language. With OpenAI, open Languages and select the languages you speak, such as Chinese and English. With DeepInfra, choose one Language, or keep Automatic for mixed speech. Its API documents a single language parameter; its effect on hosted Qwen3-ASR has not been verified with a real key. SiliconFlow stays automatic because its API does not document a language parameter. Selections belong to each service and take effect on Save.

Custom request fields

After entering a valid Custom URL and model, use Advanced Request Fields to add a field, open its row, and edit its name and values. Swipe a field or value to delete it. For example, add languages[] with two values, zh and en, or language with one value, zh, according to the service's documentation.

Names are sent exactly as entered. Several values repeat the same multipart field name; Heeler does not add []. Numbers and booleans are entered as text. file and model are reserved. Duplicate field entries, invalid names, and conflicting language / languages fields prevent Save and show a reason.

A Custom prompt replaces the shared Prompt, and response_format replaces the default json. Empty string values are allowed. The response still needs to be JSON with a text field, or text/plain; streaming responses are not supported. These are form fields, not HTTP headers. Keep authentication in API Key. The service decides which fields it accepts and uses.

Fields are saved per normalized URL and model. Editing the URL or model opens that destination's own fields; a new destination starts empty. Back or Cancel without Save leaves the saved configuration unchanged. On upgrade, an enabled legacy Send Language becomes a language field containing the device's then-current first language. Later system language changes do not change it.

Keywords and Prompt

In Settings > Voice Input, two fields shared across all services stay when you switch Service:

• Keywords: comma-separated names and terms you say, such as herdr, Claude Code, Codex, pnpm, SwiftUI, so the service spells them right.
• Prompt: context for what you talk about, sent as written.

OpenAI reads Keywords directly and leaves Prompt unchanged. DeepInfra, SiliconFlow (China), and Custom do not read Keywords; editing Keywords under them or switching to one syncs a Vocabulary: … line in Prompt so Prompt shows what is sent.

Choose a microphone

On iOS 26 and later, AirPods with high-quality recording are used automatically; other Bluetooth headphones use your iPhone or iPad microphone.

Turn on Always Use iPhone Microphone (or Always Use iPad Microphone on iPad) in Settings > Voice Input to always use the built-in microphone. It is off by default.

When something goes wrong

When a recording produces no text, the Composer shows the reason with Retry (sends the same recording again), Voice Input Settings, Open Settings (for microphone permission), and Dismiss.

• The speech service rejected the API key. — Tap Voice Input Settings, save a valid key, then tap Retry.
• Heeler can't use the microphone. Turn it on in Settings. — Tap Open Settings and allow microphone access.
• The recording was too short. — Speak for at least half a second.

Privacy

Recordings are uploaded to your chosen service under its terms and billed to your key. Heeler's developer receives none of them. See the Voice Input section of the Privacy Policy.

Last checked on 2026-11-06.