On your device
Parakeet for speech, rules and local models for text. The tested scope refers to the S24 Ultra.
Consierge in detail
An assistant that adapts to you.
Dictate, express, reply. With local models, your own style and clear choices about which context is used.
01 / Dictation
Speak without losing your train of thought.
Tap to start dictation; holding lets you record and finish in one gesture. The waveform makes recording visible. The X discards it, including by dragging and releasing over the cancel area.
Parakeet recognises speech locally. Afterwards you can use fast rules, Spokenly or Qwen for post-processing. Free-form model generation takes different amounts of time depending on the device and task.
- Dictation in enabled, suitable app text fields
- Outlook subject and message tested on the S24
- Adjustable button size, opacity, position and pause
02 / Writing style
Casual here. Business there. Always you.
Six writing styles can be assigned to apps or categories. A general profile and explicit app permissions extend dictation to other suitable fields.
Optional local style learning uses only enabled sources. Individual habits, an app context or the entire profile can be removed.
Your dictionary and snippets
Save names, brand terms and your own spellings. Text snippets expand a spoken shortcut into saved text. Search helps you find and edit entries.
03 / Replies
A suggestion. The last word is yours.
Suggested replies refer to an explicitly selected message in supported messengers. A short spoken instruction can guide the draft. You can edit the preview.
Neutral and Warm are initially enabled. Additional tones — Romantic, Flirty and Heated — are explicitly enabled when needed. Defaults can apply globally or per messenger.
Consierge prepares drafts. It does not send messages on its own or universally identify people by name.
04 / Memory
Keep context. Keep control.
Conversation memory is separate from style learning and is off by default. It stores limited, enabled excerpts encrypted. You choose a retention period of 1, 7 or 30 days.
Summaries are prepared locally and accepted only after your review. You explicitly enter key notes yourself. Short-term context, summaries, key notes or the entire profile can be removed selectively.
What does the responder do?
The experimental responder needs global and profile-specific permission. Its first automatic association is limited to an open WhatsApp text field with a visible international number, uniquely bound profile and empty draft. It prepares suggestions. Inserting and sending remain your actions. Real-chat acceptance is still pending. In Telegram and Signal, memory is used only after explicit profile selection.
Open paths forward
From your device to your own AI infrastructure.
Local first remains the principle. Extra computing power is intended to be connected deliberately. These paths are in development or planned.
Ollama
Use more powerful models on your own local network. Protected access, server discovery, model selection and fallbacks are being prepared.
Official documentation ↗LiteLLM
Your own model providers, company models and routing through your own gateway. Which data leaves the device depends on the selected route.
Official documentation ↗None of the server connections is currently verified end to end in Android. No availability date or performance commitment is promised.
Transparent development status
Real screens. Honest limits.
The screens shown come from the native Android prototype on a Galaxy S24 Ultra. The memory profile is synthetic. Other platforms are planned. Model responses can be wrong; review them before accepting.
Go to beta downloads →