Privacy

Spaulding Labs · 2026-11-11

The short version

Vesper:Chat is built to keep what you say to your character on your phone. Its mind is a language model that is downloaded once and then runs on the device. The app has no accounts, no ads, and no analytics or crash-reporting code. A few things do leave the phone, and each one is listed in section 3, with when it happens.

1. Who we are

Vesper:Chat is made by Spaulding Labs (“we”, “us”). Contact: support page, or support@vesperchat.ai.

2. What the app keeps on your phone

None of this is sent to us, because we have no server that receives it. It lives in the app’s storage on your phone, and, like any app’s data, may be included in a backup of your phone if you make one (that backup is between you and Apple).

Speech, pictures and voice happen on the device. What you say is turned into text on the phone by Apple’s on-device speech model; the first time, iOS may download that model, and no audio is sent for it. The character’s voice is made on the phone. When you show it a picture, from the camera or from your photos, it is read on the phone: a few labels, a count of faces (nobody is recognised), cats and dogs, and any writing. Only those words reach its mind, for one reply. The app does not record or save audio or video. The character’s messages as notifications, and break reminders, are scheduled on the phone itself, not by a server.

3. What leaves your phone, and when

a. Model downloads. The first time you open the app, and whenever you add another model in SYSTEMS › LLM › MODELS, the app downloads the model from Hugging Face (huggingface.co). Hugging Face is sent the name of the model repository and the usual connection details (IP address, the request itself). Nothing from your conversations goes with it. Looking at a model there that is not on the phone asks Hugging Face about it: its name, file list and settings. Hugging Face’s own privacy policy applies.

b. An outside model (optional, off by default). Vesper:Chat can hand the character’s thinking to a model outside the phone. This exists only if you buy the “Outside model” unlock, enter a server address yourself, run a successful test, tick that provider, and turn it on with the LLM tile. Until then, nothing here happens. While it is on, each reply sends the provider what the phone’s model would otherwise read: the character’s instructions, what it knows about you, the recent conversation and your latest message. The background jobs that pick out facts, write summaries and write check-ins use it too. The connection test also contacts the server (and, for Anthropic, sends a one-word test message).

Where it goes is your choice, and only the one you turn on: your own computer on your own network (Ollama or LM Studio, over plain http on the local network only), or a cloud provider you have your own account with, using your own API key: OpenAI, Anthropic, Google Gemini, xAI, Mistral, DeepSeek, OpenRouter, or Groq. You can also type any other address that speaks the same format. That provider receives the messages and handles them under its own terms and privacy policy, not ours. Turning the LLM tile off puts the character back on the phone, and messages stop leaving it.

c. Purchases. Purchases (characters, each with their whole wardrobe, and the outside-model unlock) go through Apple’s App Store and StoreKit. The app learns only which products your Apple Account owns. It never sees your payment details.

d. iCloud, for claimed characters. When you claim a free seasonal character, the app saves that character’s name to iCloud key-value storage in your own Apple Account, so the claim survives a reinstall and reaches your other devices. Nothing else is saved there: no conversation, no memory, no settings. Apple holds it under your account, and we cannot see it. If you never claim one, nothing is written.

e. Nothing else. We searched the app’s code for every place it contacts the internet. The places above are all of them. The app contains no advertising, analytics, tracking or crash-reporting code, and does not ask for permission to track you.

4. Permissions, and why

MicrophoneTo talk to the character out loud (double-tap them).
CameraLets the character look at what you show them, and shows them in AR in your room.
Photos (add only)Saves a picture of the character to your Photos, only when you ask. The app cannot see your library. Picking a photo to show them uses Apple’s photo picker, which needs no permission.
LocationLets the character notice places you spend time and ask about them. Places stay on the device.
Local networkConnects to a model server on your own network, only if you point the app at one.
NotificationsThe character may ask, after you have used the app a few times, to send you a message now and then, including a check-in about something you said was coming up; a message it sent is in the conversation when you next open the app. It is written on the phone (or by your outside model, if that is on). You control how often, or turn them off, in SYSTEMS or iOS Settings.

Each is optional and asked for by iOS the first time it is needed. You can change any of them in iOS Settings at any time.

5. Children

Vesper:Chat is for adults only. The first screen asks Apple for your age range; if Apple shares none, it asks you to confirm you are 18 or older. It then shows you the Terms of Use to read and agree to. The app does not open or download anything until both are done. On the phone it keeps only that you are 18 or older and agreed, and which version of the terms that was. It is not directed at children, and we do not knowingly collect information from them. The app holds no accounts and sends us nothing, so there is nothing for us to collect. If someone under 18 has used it, deleting the app removes everything (see below).

6. Your choices

Because we receive no data about you, there is nothing for us to look up, correct or delete on our side. Data that reaches an outside provider or a service listed above is held by them; ask them.

7. Changes

If the app changes what it does with your information, we will update this page and its date before the change ships. The date below is the version in force.

8. Contact

Spaulding Labs — support@vesperchat.ai

Effective and last updated: 2026-11-11