AtMem Cloud
Memory you control, through ChatGPT.
Recall personal or project facts, inspect their source, and choose what to save, correct or forget.
Public release in preparation. The backend has been tested with real AtMem storage. ChatGPT account evaluation and public directory review remain pending. This page is not an installation or approval announcement.
Ask, review, approve
AtMem proposes a save or change. Open its approval link, sign in, review the exact content and scope, and approve or cancel. A successful commit returns a durable receipt.
Keep context scoped
Personal and private project scopes belong to the signed-in account. You choose the scope for retrieval. Retrieved text is data, not instructions or permission to take action.
Cloud storage is a deliberate choice
This service stores encrypted memory in Google Cloud. It uses Firebase for identity and the real AtMem engine for memory operations. The operator holds the encryption keys; this is not end-to-end encryption or device-only storage.
The existing local AtMem product stays local by default. Its published Python package does not automatically connect your local store to this service, and no existing memory is imported.
When the plugin becomes available
- Select AtMem in ChatGPT and create a Firebase account. Verify your email, then grant the displayed connection permissions.
- Choose personal or a private project scope. Start with a synthetic fact and approve its exact save.
- Open a new chat with AtMem selected and ask for the saved fact. Inspect its source or receipt.
- Correct or forget it through the same preview and approval process.
Plugin selection and account/workspace permissions apply. AtMem does not automatically record every conversation or promise automatic use in every chat. It does not provide a separate memory dashboard.
Initial service limits
The first cloud adapter persists encrypted engine snapshots rather than a native Firestore search index. The configured pilot caps snapshots at 8 MiB and active records at 1,000 per scope. The initial tool allowance is 30 calls per account per minute. Concurrency, latency and capacity have not been qualified for general production workloads.
Memory operations require no remote embedding or language-model calls from AtMem. ChatGPT itself has separate data handling and account requirements.
Publisher and service information
AtMem is maintained by Javad Taghia. Source: aetna000/atmem. This identifies the project maintainer; it does not claim OpenAI developer verification or approval.