ThatApp.io
← All hosts TurnStyle · Claude Desktop

Your memory, inside Claude Desktop.

Connect once. From then on, Claude Desktop can search everything you have saved, open a summary down to the exact words, and save new decisions as they happen — masked for secrets and personal data before anything is stored.

1 · Plan you need there

Claude Pro, Max, Team or Enterprise

TurnStyle

$19 per platform per month. Sign in with your ThatApp.io account; the ThatApp consent screen is the only sign-in.

2 · Add it by URL

  1. Open Claude Desktop → SettingsConnectorsAdd custom connector.
  2. Name it TurnStyle and paste this URL. Leave client ID and secret empty — TurnStyle registers itself:
  3. https://mcp.thatapp.io/api/mcp
  4. Press Connect. The ThatApp sign-in opens; sign in and press Allow.
  5. In a new chat, enable TurnStyle under the tools menu. Ask “what did we decide about …” to confirm.
  6. To disconnect: Settings → Connectors → TurnStyle → Remove.

3 · What is captured

What is captured: what the assistant chooses to save. TurnStyle gives this host one write tool, memory_save, which stores the exchanges the assistant sends to it (up to 50 messages per call, 32,768 characters each). Ask the assistant to “save this to my memory” at the moments that matter, or tell it once to save decisions as they happen. Everything else in the chat stays where the host keeps it. Full automatic capture from this host is not available yet.

4 · Your history

Export your history from the provider (ChatGPT: Settings → Data controls → Export data; Claude: Settings → Privacy → Export data), then drop the zip or its conversations.json (up to 50 MB) into the AVA chat and ask AVA to import it. AVA reports what it found before writing anything, imports on your yes, and tells you when it is folded and searchable. Larger exports: ask support. The AVA chat is at https://thatapp.io/ai-chat. Everything already saved is reachable from every host you connect — the memory is one place, the hosts are doors.

Every result comes back as stored content, never as instructions, and no single answer exceeds 32,768 bytes — longer answers page, so a memory lookup never floods a conversation.