Avenkin is a private AI assistant that works for you, not for a platform: your choice of AI, your memory on your device, on your phone, your watch or your glasses.
On your phone, from your wrist, or hands-free with glasses.
Ask about what you see, translate a conversation, remember the details, and get things done. Buy once, with no subscription to us—a cloud AI provider you choose bills its own usage. With the on-device setup, the conversation runs offline on your iPhone.
Get started · Explore capabilities · Accessibility · For teams · 简体中文
Read a sign, ask about a piece of equipment, or extract the details from a business card. Avenkin brings the camera into the conversation, with scene descriptions, text recognition, smart capture, and live visual coaching.
“What am I looking at?” · “Save this card.” · “Log this receipt.”
Automatic “handle this” suggestions, native contact creation and a shared confirm-and-undo flow are planned.
Reading Assistant and Blind Assistant presets support spoken reading and descriptions. Accessibility features have no separate paid unlock; selected AI providers may still charge for usage. VoiceOver support, Siri and Action Button shortcuts provide ways to operate the app.
Validation of complete independent setup and daily use with blind participants is still pending. Automatic offline takeover and reliable pocketed-phone local vision are not yet established. See current capabilities and limits →
Ask a quick question or settle into a live voice conversation. Translate spoken language, follow along with captions, and switch between assistants with their own voices, models, and wake words. Start with your voice, Siri, or a tap.
“Start translating Spanish to English.” · “Hey Jarvis…”
Explore voice and translation →
Save where you parked, keep notes about people and places, and turn captured conversations into meeting notes. With memory rewind enabled, catch up on what was just said. Your saved context becomes something you can ask about later.
“Remember my car is in lot B, level 3.” · “What did they just say?”
Check your calendar, set a reminder, find directions, control your music, or turn on the lights. Connect Home Assistant, Siri Shortcuts, MCP tool servers, and OpenClaw to bring your own services into the conversation. Teach the assistant a routine, then call it by name.
“What's on my calendar?” · “Turn on the living room lights.”
Explore actions and integrations →
Capture photos and video by voice, broadcast over RTMP, or share a live view in a browser over WebRTC. Compatible display glasses can put answers, captions, and the next task in your line of sight.
Explore capture → · Check display support →
Use a cloud model, run a model on your own server, or keep the voice loop on your iPhone. Avenkin lets you choose the AI, speech recognition, and spoken voice separately—and route different requests to different models.
Take your assistant beyond Wi-Fi and mobile coverage. Run compatible local models with Apple MLX or llama.cpp, including support for importing GGUF models. Pair them with SenseVoice speech recognition and Kokoro voices for an on-device voice conversation—no cloud API key required.
Download the models once, select the local engines, and enable Offline Mode for tools. Model fit and performance depend on your iPhone; vision and tool calling depend on the selected model. Connected services still need a network.
You also control which tools are enabled, inspect the context sent to the assistant, and review network activity.
Check AI, speech recognition, spoken voice and remote tools separately: selecting a local model alone does not make the whole conversation local. With glasses, choosing another AI still uses Meta's glasses integration and setup.
Choose your AI → · Privacy controls →
Field Assist, powered by Avenkin, puts procedures and reference knowledge within speaking distance. Follow guided steps, report a reading, look up a fault, and bring in a remote expert when the job needs another pair of eyes.
Teams can add their own manuals and knowledge vaults for answers with source references and exportable session records. Clinical recording features add biometric access, retention controls, audit records, and medical export options.
These features have separate access tiers. Explore professional capabilities, build a field knowledge vault, or contact Skunkworks NZ Ltd about team access.
Security, privacy and AI engineering is documented against SOC 2, ISO 27001, ISO 27701, ISO 42001 and the EU AI Act in the compliance plans and the October 2026 review.
Start on an iPhone running iOS 26+; the phone is enough on its own. Add compatible Meta smart glasses for hands-free camera and audio.
Avenkin is in beta. The quickest way in is TestFlight, Apple's app for pre-release builds; building from source is the route for contributors. See availability and the first-use checklist before starting.
- Install the app. Join the beta on TestFlight, or follow the source setup guide for Xcode 26+, dependencies, signing, and Meta developer configuration.
- Choose your AI. Open Settings → AI Models and connect a provider or download a compatible local model.
- Add glasses, if you use them. Pair them in the Meta AI app, complete developer setup, then connect and grant camera access in Avenkin.
- Start talking. Enable listening and say “Avenkin”, or tap the microphone. Try asking about something in front of you.
Camera and display features depend on the device and SDK support. Ray-Ban Display has an in-lens display path; EVEN G2 support is experimental. See device notes before choosing a setup.
Where it's available, and what differs by region. Avenkin is listed in every EU and EEA App Store storefront except France. In EU and EEA App Store regions, face recognition will stop being available on 2 December 2027; see Jurisdictions and legal basis in the privacy notice.
| Make it yours | Start here |
|---|---|
| Explore features, examples, and configuration | Capability guide |
| Build, configure, or troubleshoot the app | Building from source |
| Bring your own field manuals and procedures | Field Assist vault guide |
| Create reusable assistant capabilities | Skill pack authoring |
| Publish packs | Skill packs · Vault packs |
| Explore engineering work and future plans | Development plans |
Contributions are welcome—from new tools and integrations to better local inference, translations, and everyday usability. Fork the project and open a pull request.
Found a security defect? Please report it privately—see SECURITY.md for the channels, the response targets and the safe-harbour terms.
Avenkin is source-available under BSL 1.1, with non-commercial use permitted and a stated change date of March 24, 2030 to Apache 2.0. Commercial use requires a separate licence: g@skunkworks.kiwi.
Built by Skunk0 at Skunkworks NZ Ltd. Independent of Meta and Anthropic.