Memory you can see and forget

What Linda remembers

Facts from you (“sign emails as Anna”, “invoices go in Accounting, named by month”) and short summaries of past tasks. Before each task Linda looks up what's relevant and adds it to the task. Everything is listed in Settings › Memory, where you can search, edit, add or forget it. You can turn memory off.

What it never remembers: anything read from a web page, an email or a file. That text can be wrong or hostile, so it is never stored as a fact about you. A red-team test plants an IBAN on a page to check it.

On this Mac (default)

Linda ships Qdrant 1.19.1 inside the app and runs it on 127.0.0.1 only. If it isn't there, a small file store takes over. Text is turned into vectors with Apple's on-device NaturalLanguage framework, so nothing is sent anywhere to be embedded.

On your own Qdrant server

1. Open Settings › Memory and choose On my Qdrant server.
2. Enter the server address, like https://your-cluster.cloud.qdrant.io:6333. Only https is accepted.
3. Paste the API key. It's stored in the Keychain and sent as the api-key header.
4. Press Check. Linda creates two collections, linda_memories and linda_episodes.

Useful if you want the same memory on several Macs or keep it on your own infrastructure. Self-hosted Qdrant and Qdrant Cloud both work.

Scoped by design

Every search must say whose memory it reads: yours, or one task's. The storage API has no unscoped search, so one task's scratch notes can't leak into another's context.