Your archive may hold many parts of your life. A tool helping with one task should receive the part you choose for that task.
SEED's connections are built around a clear selection: who receives information, what they may access, what they may do and how long access lasts. This guide describes those choices for the launch experience. Access opens through the SEED waitlist.
Collecting data and sharing data are separate choices
| Choice | What it allows |
|---|---|
| Read a public download guide | Get instructions for exporting records from a service. No access to your private account is needed. |
| Connect a source to SEED | Let an available source connection collect the information shown in its permission flow. |
| Upload an archive | Put the records you selected into your SEED vault. |
| Connect an assistant or application | Let that recipient request the SEED material and operations you approve. |
Collecting records into your vault does not automatically share them with ChatGPT or another assistant. A tool's connection to SEED does not give it the passwords or provider credentials used to collect your records.
Review the access before you grant it
A useful permission choice should answer five questions:
- Who is receiving access? Check the application and the SEED account you are connecting.
- What is included? Review the selected records, brief or collection and the information it contains.
- What can the tool do? Reading context is different from saving a proposed memory or changing a record.
- How long does it last? Choose from the durations supported by that connection.
- What happens to new information? For ongoing access, check whether additions or changes to the selected collection become available.
A familiar application name is not a reason to approve every request. Use the connection flow linked from SEED or the verified client instructions, and check the selection shown in SEED.
Choose one handoff or ongoing access
A reviewed handoff
Select a brief or fixed set of records for a particular task. For example, share your approved chapter outline and the passages you want a writing assistant to use.
Review the material before sharing. An expiry can end future retrieval of that selection, but it cannot erase a copy the recipient has already received.
A selected collection
Give a tool ongoing access to a defined collection, such as one project's notes. This is useful when you return to the same work across several conversations.
Check the collection's contents and its rules for new or updated records. Access to one collection does not imply access to other projects, private messages or health records.
If a tool needs something outside the selection, that requires another access decision. A tool cannot make the selection larger simply by asking a broader question.
See sources and correct the record
Useful context should show where it came from and when it was recorded. A past preference, an imported statement and an interpretation are not always the same thing.
Review source details when an answer matters. Correct outdated or mistaken information in SEED so future retrieval can reflect the correction. A connected assistant's suggestion should remain identifiable as a suggestion until the appropriate review step is complete.
Permission to read your history does not authorize a tool to rewrite original records, impersonate you, publish in your name or reproduce your voice. Those are different actions with different requirements.
Understand what leaves the vault
SEED operates the core intelligence that organizes and interprets your vault on its private infrastructure. When you use a connection, the approved response is sent to the external assistant or application.
That recipient may retain the response in a conversation, a document or another record according to its terms and your settings. The fact that your originals remain in SEED does not change how the recipient handles the information it receives.
Review the recipient's privacy and retention terms before sharing. A short, relevant selection is often more useful than a large personal archive and easier to understand before you disclose it.
Some information cannot use every connection
Your permission matters, and so do the rules of the source and destination. The same record may be suitable for one supported workflow and unavailable in another.
For example, the public ChatGPT plugin must exclude payment-card data, protected health information, government identifiers and authentication secrets under OpenAI's rules. Other sensitive information can require additional conditions. A general memory connection should not be used as a route to bypass those restrictions.
Records collected from a provider may also carry limits on reuse and transfer. Summarizing a record does not automatically remove those limits. SEED's supported connection flow must account for the selected material and destination.
For health records or identity-related tasks, use a separately supported and appropriate workflow. Do not paste credentials, recovery codes or identification documents into an integration's setup fields.
End access, disconnect a source or delete records
These controls do different things:
| Action | What it means |
|---|---|
| End a tool's access to SEED | Stop that connection from retrieving more information. It does not recall copies already sent. |
| Disconnect a source | Stop further collection through that source connection. Check the vault's retention controls for records already collected. |
| Delete records from SEED | Use the vault's deletion process for the selected stored records. This is separate from disconnecting a source or recipient. |
| Delete a conversation in another service | Follow that service's own deletion controls and terms. SEED cannot perform or guarantee its deletion merely by revoking access. |
If you shared more than intended, end future access first, then review what was disclosed and the recipient's available deletion controls. Keep a copy of records you want to preserve before requesting their deletion from your vault.
Use the right account
Personal and work records can have different owners and sharing rules. Check the active account when connecting, especially if an assistant or application supports several accounts or workspaces.
Only include records you are entitled to share for the task. A file being present in your archive does not necessarily give you permission to disclose everyone else's information in it.
When access expires or a connection needs to be renewed, check the account and selection again. Renewing a connection should be an understandable choice, not a reason to grant unrelated access.
Questions to ask before connecting
- Can I name the collection or brief this tool needs?
- Do I understand what new information could become available during an ongoing connection?
- Does the receiving tool need to read, save a proposal, or do something more?
- Am I comfortable with the recipient keeping the information under its terms?
- Do I know how to stop future access?
You can begin with public preservation guidance and decide later whether to connect your vault. Read the ChatGPT guide, explore MCP for agents or learn about the developer API.
Official references
Checked on 28 October 2026. These references explain the standards and provider policies relevant to sharing your data.
- OpenAI plugin guidelines set the data and permission conditions for the public ChatGPT integration.
- Google Workspace user-data policy and Limited Use FAQ explain restrictions relevant to information collected through Google scopes, including derived information.
- MCP authorization specification describes the connection's authorization protocol. SEED's selected-material controls are a product responsibility in addition to that protocol.