Dawn — privacy & data handling

What conversation data Dawn stores, how it's encrypted, what we share with the AI provider, and your rights.

Quick answer

Conversations are encrypted at rest with per-user keys, never used for AI training, and forwarded to the AI provider with identifying fields stripped. Full export, full deletion. Crisis-detection flags are stored locally for safety only.

privacydatasecurity

ShareX LinkedIn

Dawn handles some of the most sensitive content in the product. The data handling is conservative on purpose; the explainer below is meant to be specific enough that you can verify the claims rather than trust them.

What gets stored

Per Dawn conversation:

Your messages and Dawn's responses (full text).
Timestamp of each message.
Sentiment scores computed from your messages.
Crisis-detection flags (when a conversation triggers them — see Crisis detection).
The conversation title (auto-generated; you can edit it).

Per account-wide companion memory:

Up to 50 ranked facts (see Companion memory).
Source conversation reference for each fact.

What is not stored

Your location (we don't request or use it).
Device identifiers beyond standard analytics (browser version, OS).
Data from other apps on your device.
Biometric data of any kind.

Encryption

Surface	At rest	In transit
Messages + responses	AES-256, per-user key	TLS 1.3
Companion memory	AES-256, per-user key	TLS 1.3
Sentiment scores	AES-256	TLS 1.3
Crisis-detection flags	AES-256, separate key tier	TLS 1.3

Per-user keys mean each user's conversation data is encrypted with a key tied to their account. There is no single shared key.

What goes to the AI provider

Dawn uses an external AI provider (currently Anthropic / OpenAI per our enterprise agreements) to generate responses. When you send a message, here's what happens:

Daybreak strips identifying fields

Your name, email, account ID, and any auth tokens are removed from the request before it leaves our servers.

The provider receives the message text

Plus a temporary session ID and the conversation history Dawn needs for context.

The provider generates a response

Per our enterprise agreements, the provider:

Does not retain message text past 30 days.
Does not use your data for model training.
Is contractually bound to delete data on request.

Daybreak stores the response

Encrypted, tied back to your account, ready for the next turn.

The provider sees text — they don't see you. From their side, each request is an anonymous session.

Data retention

Data	Retention
Active conversations	As long as your account is active
Deleted conversations	Permanently removed within 24 hours
Crisis-detection flags	Same retention as the conversation
Provider-side request data	≤ 30 days, then deleted
Account deletion	All Dawn data purged within 30 days

Backup rotation completes the deletion within 60 days for backup tiers.

Your rights

Export. All Dawn conversations and companion memories are in your data export under dawnConversations[] and companionMemories[].
Delete individual conversations. From Settings → Dawn AI → Conversation history. Permanent.
Delete companion memories. From Settings → Dawn AI → Companion memory. Permanent.
Delete all Dawn data. Same settings page, Delete all Dawn data button. Removes conversations + memories without deleting the account.
Delete the entire account. See Delete account.

Crisis-detection flags

When Dawn's classifier identifies a conversation as crisis-adjacent (suicidal ideation, self-harm, severe distress), a flag is stored alongside the conversation. The flag's purpose is exclusively safety:

It does not change billing.
It does not affect account standing.
It is never shared with third parties (including law enforcement, insurers, family).
It is included in your data export so you can see what's been flagged.
It is deleted when you delete the conversation or your account.

See Crisis detection for what the flag actually does at the moment it fires.

Common pitfalls

Sharing identifying details in messages

We strip identifying fields from the metadata. We don't strip them from the message text. If you tell Dawn "my name is Sam, I work at Acme", that text is what the provider sees. If identification matters to you, talk to Dawn at the level of what's happening rather than who you are.