Connect STEM shared memory to ChatGPT

Documentation reviewed 2026-10-25. Host verification dates are listed separately.

Tested workflow. A published ChatGPT STEM connection has completed OAuth, an explicitly authorized save, recall from another connected assistant and recall in a fresh chat. This is workflow evidence, not a guarantee for every account, model or device.

Connect through the published listing

1. Open the Stem by Awase LLC listing in ChatGPT and choose its installation or connection action. You can also start from STEM's Connect screen.
2. Sign in to the STEM account you want your assistants to share. Confirm that account before approving access. Choose Only save when I ask for a manual first test, then complete OAuth approval and the return to ChatGPT in the browser flow that initiated the connection.
3. Start a conversation with STEM enabled or selected, then run the save and recall test. Keep any normal ChatGPT tool approval prompts in place.

Use the same STEM account in each assistant. Your assistant may have its own subscription requirements, account restrictions and tool approvals. A connection is ready when an authenticated STEM tool succeeds.

Verify persistent memory across conversations

Start with manual saving: choose Only save when I ask during STEM consent if that choice is shown. Follow the three-step save and recall check: explicitly approve one harmless note, confirm the successful save result and memory ID, then recall that note in a fresh conversation or another connected assistant. Compare the retrieved text and ID before using it. An opened link, installed configuration or assistant saying “connected” is not proof of a completed save.

Bring context from another assistant

If Claude or Gemini has already saved a project note to this STEM account, ask ChatGPT to recall that project from STEM. ChatGPT's own memory and chat history are separate. Connecting STEM does not import them or combine accounts. Use the Claude-to-ChatGPT example to bring a reviewed note into ChatGPT, or the ChatGPT-to-Claude workflow and observed native recall for the reverse direction. See context transfer for import and synchronization limits.

If authorization is waiting for a callback

Email verification and connection approval are separate steps. In the sign-in window opened by this client, confirm the intended STEM account, finish the displayed approval and memory choice, and follow Continue to your AI through the return to the original client. Keep the original client page or app open. Check its connection state and available tools after returning.

If the window expired, a button no longer responds or the client still reports no credentials, start a fresh authorization from that client’s connection settings and use the newest sign-in flow. Finish that fresh attempt rather than returning to an expired approval screen. Never paste passwords, email codes, bearer tokens or private keys into a conversation. Check an existing working connection before creating a duplicate.

Choose what gets remembered

Ask explicitly to save with remember. Optional automatic memory is a separate choice for each eligible OAuth connection in STEM Settings → Memory preferences. The agent must call memory_status with {} before an automatic save, require automaticSavingAllowed: true, and use remember_automatically. Host approvals still apply. Switching to manual does not erase existing memories.

STEM saves selected context supplied by a connected assistant. Connecting does not import earlier conversations, give access to other chats, or guarantee a save on every turn. Do not save passwords, tokens or secrets.

Agents and developers can inspect the server reference and tool schemas. The ChatGPT listing is the connection destination; availability is controlled by ChatGPT.