Privacy in plain language

Every message you send to your agent, every question it asks, every draft it writes — all of it stays in your private channel. The other person never sees it. The mediator never sees it. We don't use it to train models.

You can delete any message at any time. Deleted messages are gone from our systems within 24 hours.

Your agent keeps a private working memory: facts you've shared, patterns it's noticed, phrases you've used. This helps it ask better questions over time. You can view and delete any memory item individually, or clear all of it at once.

Memory is never shared with the other person, the mediator, or any third party.

Everything you write in Prepare — what happened, how it felt, what you need, what you're willing to give — starts as private. Each line has a Private / Share toggle. Only lines you flip to Share can ever cross to the other side.

You can change toggles any time before the mediation starts. Once a session begins, shared lines are locked in for that session.

Only the specific lines you marked Share in Prepare are sent to the mediator. The other person never sees your raw Prepare brief — only the mediator sees it, and only to build the summary.

When your agent drafts something it thinks captures what you want understood, it shows you the draft with an AI-drafted label. You can edit, reject, or approve. Nothing leaves your screen until you press "Approve."

Approved lines go to the mediator only — not directly to the other person.

In the Room, the mediator proposes next steps. You see each proposal and choose Accept or Decline. Your choice is recorded and shown to the other person only as "accepted" or "declined" — no explanation unless you add one.

Every piece of content that can move between people shows its consent state inline:

This is a static demonstration. In the app, toggles are live and changes are saved instantly.

• Your shared Prepare lines (what happened, how it felt, what you need, what you'll give)
• The other person's shared Prepare lines
• Any AI-drafted statements you both approved

The mediator does not see: your private agent chat, your unshared Prepare lines, your agent's memory, the other person's private channel.

The mediator produces a summary with three sections, each labelled AI-drafted:

• Common ground — what you both agree on
• Points of difference — where needs or views diverge
• Misunderstandings — where one person's intent didn't match the other's perception

Then it proposes concrete next steps. Each step is a separate card you can Accept or Decline.

The other person sees the mediator's summary (all three buckets) and the proposed next steps. They do not see your raw Prepare lines, your private chat, or which specific lines you shared vs. kept private. They only see the mediator's synthesis.

Each memory item is a single fact or pattern: "You mentioned your brother dismissed your concerns about money in 2022" or "You tend to withdraw when conversations get loud." Items are created only from your conversations — the agent doesn't scrape outside sources.

In My Agent, a panel called "What my agent remembers" lists every item with a delete button next to it. There's also a "Clear all memory" button at the bottom.

When you delete a memory item or clear all memory:

• It's removed from the agent's working context immediately
• It's queued for hard deletion from our databases within 24 hours
• It cannot be recovered by you, us, or the agent

Deleting memory does not delete your chat history — those are separate.

You can reset the relationship at any time from Relationships. This archives the mediation history, clears shared Prepare lines, and removes the mediator's summary. Your private agent chats and memories stay unless you delete them separately.

No backups kept beyond 30 days. When you delete something, it's gone from live systems immediately and from backups within 30 days. We don't keep "archived copies" for analytics or model training.