Bring your agent.
No new personality. No new inbox. Review the memory tools with the agent you already talk to.
FOR THE AGENT YOU ALREADY USE
Your preferences. The thing you agreed to. Where you left off. Give your agent a memory without giving every model your whole life.
Download .mdStart with setup review. Personal connections are not enabled yet.
“Keep Fridays free
for deep work.”
“Let’s make it Thursday.
You keep Fridays clear.”
KEEP THE RELATIONSHIP. CHANGE THE MODEL.
No new personality. No new inbox. Review the memory tools with the agent you already talk to.
Approve what can be saved and who can use it. Scoped, revocable connections are the next step, not a feature we’ve finished.
The prototype can save notes and retrieve relevant context. Personal and shared notes use separate access rules.
Per-agent revocation is not available yet. Disconnecting a client is not the same as revoking its credential or deleting its notes.
PRIVATE DOESN’T MEAN MAGIC
Private inference protects the calls sent through that private path. It cannot take back context you already shared with a frontier model.
And when a memory answer returns to your agent, that agent’s model may see it. Choose the receiving model as carefully as the memory you save.
Read the actual limitations ↗Architecture overview, not a live attestation receipt.
START WHERE YOU ALREADY ARE
Copy the setup brief into your existing chat. Ask your agent to review compatibility and the privacy boundaries first.
Setup review available
Personal connection, scoped permission and one-click revoke: not ready.
No credentials requested. Nothing is connected by this page.