Your keys, your provider
You supply your own API keys. They are stored in the iOS Keychain on your device, and the app talks straight to the provider you already trust and already pay. Nothing is proxied through a bpaas.it server.
Coming soon to iOS Proof of concept
BPaaS Agent
A voice agent that runs on your keys — or on nothing but your phone. Speech in, an answer out loud, and every part of it is yours to choose.
Bring your own credentials
Most assistants ask you to trust a new company with your conversations. This one doesn't ask you to trust us at all.
Or no provider at all
Choose local models and the whole pipeline runs on the phone — speech recognition, the language model, and the voice. Audio, transcripts, and inference never leave the device. It works in airplane mode.
Even the phone calls
Call features run through your own Twilio account, with call tokens created on your device using your own credentials. No third-party server sits between you and the call.
Your notes and your history
Notes, your saved profile and the activity timeline stay on the device, encrypted at rest and left out of diagnostics exports. Location is off until you switch it on, and even then the assistant is told your approximate city, never your exact position.
Named, not handed over
Your saved details reach the assistant as names, not values: it is told there is a place called Home, never the address. The real value is filled in on your phone at the moment you read it, hear it, or a command acts on it — so it can say “taking you home” without ever having been told where that is. One switch shows you exactly what it receives; another turns the whole thing off.
No account, no analytics
There is no sign-up, no telemetry and no tracking. The app talks to the providers you configured and to nothing else. Logs are redacted where they are written, so keys and personal values never reach a log file in the first place.
Mix and match per stage
Speech-to-text, the assistant, and text-to-speech are chosen independently. Run a local model for privacy and a cloud voice for quality, or any other combination that suits the moment.
Six modes, one app
Chat, Map and Notes are the same assistant seen through different screens — switching between them never interrupts listening.
Chat
Talk or type. Hands-free listening that survives the agent's own voice, a wake word so it only acts when you mean it, and commands that fire from partial speech instead of waiting for you to stop talking. It controls music, searches the web, and reports on the device's own state.
Map
A map of where you are, nearby search, and the whole assistant riding along in the copilot seat. Say “navigate to the airport” and turn-by-turn opens in Apple Maps or Google Maps — your choice, both free — while the app keeps listening in the background, so the wake word still works mid-drive for music, questions or a new destination. Once a trip is running, it can find a restaurant or a filling station along the actual route.
Notes
A place to think. Write notes, or tap Dictate and speak one word for word. Tap the lightbulb and the assistant reads what you have written and offers a few ideas that build on it, plus a question that might unblock you — only when you ask, never on a timer. Nothing you write is ever deleted automatically, and your notes stay on the device.
Interpreter
Put the phone on the table in a meeting held in another language. Every utterance appears with its translation and a badge showing the language it detected. Have it speak the translation aloud, suggest lines for you to tap, or just prompt you with the text to say yourself.
Call Translator
Call someone who does not share your language, through your own Twilio account. Four ways to run it: Translator interprets both directions live; Copilot keeps a running summary and suggests what to say, saying nothing until you approve it; Delegate makes the call for you from a script you wrote and approved, and hands back the moment you speak; Transcribe just keeps an honest written record. The far side is told out loud what is happening.
Camera
A viewfinder you can talk to. Reading text, scanning a QR code or a barcode, and counting how many people are in frame all happen on the phone, for free, with no connection — shapes, never identities. Ask “what do you see?” and that one frame goes to the cloud assistant you chose; with a local assistant nothing is sent at all. It can also watch for a while and tell you what changed, and record video with sound — and a recording never leaves the phone, because it holds other people's voices.
It does things, not just answers
Ask for something and it acts. Anything costly or hard to undo asks you first, and only you can answer — the assistant can never approve its own request.
Music
Play, pause, skip, shuffle, repeat, seek and set the volume. Ask what is playing, list your playlists, or play one by name — “pop espanol” finds Pop Español, and when two could match it asks instead of guessing. Music keeps playing around voice turns, and the lock screen, AirPods and CarPlay controls all work.
Web search
It reads the top results and gives you one answer with its sources, rather than a list of links to open yourself. If a search comes back empty it rephrases the question once and tries again. Google, OpenAI or Claude — your choice.
Places you save
Say “save this as home” and “take me home” is enough from then on. A place saved from a map result keeps its exact coordinates, so it routes with no signal at all. Add the people you phone through Apple's own picker and “call Dad” works — the app never gets access to your address book, so there is no permission to grant.
Reminders
“Remind me in six minutes the eggs are done.” Armed as real iOS notifications, so they fire even if the app is closed or the phone restarts. They can repeat — “every Monday” — and every one that arrives offers to remind you again in ten minutes. The confirmation repeats the time back to you, so a misheard time is caught before it matters.
Memory
“Remember that my locker code is 4321.” It keeps it, shows you where it came from, and has it in later conversations. Nothing is remembered unless you ask — there is no background capture — and one switch stops the assistant being shown any of it while leaving your list untouched.
Routines
Say a goal that takes several steps and it plans the commands and runs them in order, with a checklist that reports what completed, what failed and what never ran. Write your own routines and start them by name. “Stop everything” is a hard stop, and tells you exactly what it stopped.
Swappable engines
You are not tied to one vendor. Local, on-device and cloud options sit side by side for every capability, chosen in Settings and switched whenever you like — no reinstall, no account migration.
whisper.cpp llama.cpp Apple Intelligence Apple Speech & Siri voice OpenAI Claude Gemini ElevenLabs OuteTTS MCP servers
External tools connect over MCP: register a server, review every operation it offers, and enable them one at a time. By default each call asks you first and shows exactly what would be sent.
Work with us
This app is the proof. The same foundation is what we build on for clients, which is why a bespoke agent takes weeks rather than years.
That is the offer: we can build an agent for your customers, or turn your idea into one — voice-first, on your infrastructure and your credentials, with the governance and integration discipline we bring to the rest of our work.
Questions people ask
What does “bring your own credentials” mean?
You use your own API keys from the provider you already have an account with — OpenAI, Anthropic, Google or ElevenLabs. The keys are stored in the iOS Keychain on your iPhone, and requests go straight from your device to that provider. BPaaS never sees your keys or your conversations.
Does it work without an internet connection?
Yes. Assign local models and speech recognition, the language model and the voice all run on the iPhone itself. It works in airplane mode, and no audio or transcript leaves the device. Apple Intelligence can also be used as the on-device assistant on supported iPhones.
Which AI providers does it support?
For speech-to-text: whisper.cpp on device, Apple Speech, OpenAI, Gemini and ElevenLabs. For the assistant: llama.cpp on device, Apple Intelligence, OpenAI, Claude and Gemini. For speech: OuteTTS on device, Apple voices, OpenAI, Gemini and ElevenLabs. Each stage is chosen independently.
What can it do besides answer questions?
It plays music, searches the web and summarises what it read rather than handing you links, opens turn-by-turn navigation, takes and dictates notes, sets reminders that fire even when the app is closed, remembers facts you ask it to keep, looks through the camera, and runs multi-step routines you write yourself. Anything costly or hard to undo asks you first.
Can it translate a live phone call?
Yes. Call Translator places a real call through your own Twilio account and transcribes and translates both sides live, with the other party told that an AI interpreter is on the call. There is also an in-person Interpreter mode for conversations around a table.
When is it available, and what does it cost?
It is a proof of concept and not yet on the App Store. Join the waitlist and we will email you at release. Because you bring your own keys, you pay your provider directly for what you use rather than paying us a markup.
Can you build an agent for my company?
Yes — that is what this app is a proof of. We build on the same foundation for clients, so a bespoke voice agent is a matter of weeks rather than years. Request a consultation and tell us what you have in mind.
Join the waitlist
We are deciding what to build out first. Tell us what you would actually use and you will hear from us when it ships.
About your data. Your credentials and conversations never reach our servers — keys stay in the iOS Keychain and the app talks directly to the provider you chose, on your own account. Pick local models and nothing leaves your device at all. This form is separate: the email and answers you submit here are stored, so we can contact you at launch and count the responses. See our User Privacy Choices.