AI Privacy

Last updated: 2026-09-30

Introduction

This page describes how data is handled when you use AI features in Anasa. It complements the Privacy Policy, which covers the app and the account system as a whole.

1. Local-First by Default

Anasa's AI features send note content off your device only when you use an AI setup that runs in the cloud:

On-device (Apple Intelligence): On supported Apple hardware, AI features can run entirely on your device. Your content never leaves the machine.
Local models (Ollama): If you run a local model server such as Ollama, AI requests are processed on your own hardware. Your content never leaves the machine.
Managed service (Anasa AI) and bring-your-own-key (BYOK) modes send content off-device only once you enable them. After that they send the content a feature needs when you invoke it - in chat, that includes excerpts of other notes retrieved as relevant to your message - and, when the provider supports embeddings, the notes you create or edit are also embedded in the background (section 2). They are described below.

This page covers AI only. The optional Anasa Cloud sync services also send content to our servers when you switch them on; they are described in the Privacy Policy.

Cloud AI is never switched on for you. At first launch, if Apple Intelligence or a local Ollama server is available, Anasa turns that on-device option on; you can switch AI off in Settings → AI Features.

2. What Anasa AI Sends

When you use the managed Anasa AI service:

When you invoke a feature, the note text needed for that feature is transmitted - for example, the note you asked to summarize, or the messages of the chat you are having.
Chat retrieval: to answer from your own notes, a chat message is also sent with short excerpts of the notes in your vault that are most relevant to it.
Background indexing: while Anasa AI is on, notes you create or edit are also sent for embedding in the background, a few seconds after each change, so semantic search, chat retrieval and Smart Connections can find them. Rebuilding the index from Settings sends every note in the vault the same way. Anasa AI does not otherwise upload your vault or sync your notes.
Content is sent over TLS to anasa.md and forwarded through the Vercel AI Gateway to our AI model provider - currently DeepSeek for generation and OpenAI for embeddings - to process the request.
Your content is not used to train models and is not stored beyond what is needed to process the request. Anasa's servers do not retain the note text after the response is returned.
Embedding requests work the same way: the resulting vectors are returned to your device and stored locally there, not on Anasa's servers.

3. What Anasa AI Records

For credit accounting, each successful managed-AI request writes a usage record to our database containing:

The license id the request was made under.
The feature invoked (e.g. chat, summarize, suggest tags).
The credit count deducted (recorded as a negative entry when credits are refunded).
A timestamp.
Minimal request metadata used for accounting and debugging - for example, the number of messages in a chat request, or the id of the note a feature ran on. This metadata never includes your note text or note titles.

These records exist so your credit balance and usage history are correct; they are retained under the same account-data rules as the rest of your account (see the Privacy Policy).

4. Bring-Your-Own-Key (BYOK)

If you configure your own API key for a provider, requests go directly from your device to that provider. Anasa's servers are not involved, no credits are charged, and no usage records are created on our side. Data handling is governed entirely by your chosen provider's terms and privacy policy.

5. Rate Limits and Abuse Controls

To keep the managed service available and prevent abuse, Anasa AI enforces per-license rate limits and caps on request input size. These controls operate on request identifiers and sizes, not on the content of your notes, and do not involve retaining your content.

6. Contact

Questions about AI data handling: email hey@anasa.md (Entro314 Labs).

1. Local-First by Default

Anasa's AI features send note content off your device only when you use an AI setup that runs in the cloud:

On-device (Apple Intelligence): On supported Apple hardware, AI features can run entirely on your device. Your content never leaves the machine.

Local models (Ollama): If you run a local model server such as Ollama, AI requests are processed on your own hardware. Your content never leaves the machine.

Managed service (Anasa AI) and bring-your-own-key (BYOK) modes send content off-device only once you enable them. After that they send the content a feature needs when you invoke it - in chat, that includes excerpts of other notes retrieved as relevant to your message - and, when the provider supports embeddings, the notes you create or edit are also embedded in the background (section 2). They are described below.

This page covers AI only. The optional Anasa Cloud sync services also send content to our servers when you switch them on; they are described in the Privacy Policy.

2. What Anasa AI Sends

When you use the managed Anasa AI service:

When you invoke a feature, the note text needed for that feature is transmitted - for example, the note you asked to summarize, or the messages of the chat you are having.

Chat retrieval: to answer from your own notes, a chat message is also sent with short excerpts of the notes in your vault that are most relevant to it.

Background indexing: while Anasa AI is on, notes you create or edit are also sent for embedding in the background, a few seconds after each change, so semantic search, chat retrieval and Smart Connections can find them. Rebuilding the index from Settings sends every note in the vault the same way. Anasa AI does not otherwise upload your vault or sync your notes.

Content is sent over TLS to anasa.md and forwarded through the Vercel AI Gateway to our AI model provider - currently DeepSeek for generation and OpenAI for embeddings - to process the request.

Your content is not used to train models and is not stored beyond what is needed to process the request. Anasa's servers do not retain the note text after the response is returned.

Embedding requests work the same way: the resulting vectors are returned to your device and stored locally there, not on Anasa's servers.

3. What Anasa AI Records

For credit accounting, each successful managed-AI request writes a usage record to our database containing:

The license id the request was made under.

The feature invoked (e.g. chat, summarize, suggest tags).

The credit count deducted (recorded as a negative entry when credits are refunded).

A timestamp.

Minimal request metadata used for accounting and debugging - for example, the number of messages in a chat request, or the id of the note a feature ran on. This metadata never includes your note text or note titles.

These records exist so your credit balance and usage history are correct; they are retained under the same account-data rules as the rest of your account (see the Privacy Policy).

Local by default

You choose what is sent

Your key, your provider

Introduction

1. Local-First by Default

2. What Anasa AI Sends

3. What Anasa AI Records

4. Bring-Your-Own-Key (BYOK)

5. Rate Limits and Abuse Controls

6. Contact

AI Privacy

Local by default

You choose what is sent

Your key, your provider

Introduction

1. Local-First by Default

2. What Anasa AI Sends

3. What Anasa AI Records

4. Bring-Your-Own-Key (BYOK)

5. Rate Limits and Abuse Controls

6. Contact