harnyUse case
Memory for a personal assistant
A personal assistant serves one person for months. Memory for an assistant has to remember what that person said, keep it private to them, and change it when they change their mind.
What you are building
An assistant that knows each user's preferences, plans and history, and acts on them without being told again.
Why memory decides whether it works
- People repeat themselves when the assistant forgets. Each repeat costs trust.
- Preferences change. The newest statement has to win, and the old one must stop being quoted.
- Memory must stay private. What one user told the assistant is never another user's.
What Harny provides
- Memory per user: each call names the user, and recall returns only that user's memory.
- Connected accounts: bring in what each user chooses to connect, scoped to that user.
- Deletion that reaches the memories derived from a send.
Getting there
- Send what the user says and does to
POST https://api.harny.ai/v1/memory/ingest, with the user named inX-Harny-User. - Call
POST https://api.harny.ai/v1/memory/recallbefore each reply and add the evidence to the prompt. - Let the user delete what the assistant knows by deleting the sends that hold it.
Frequently asked questions
Can one user's assistant read another user's memory?
No. Every read is scoped to the user named in the call.
How does the assistant learn that a preference changed?
Send the new statement with its date. Recall returns the current value and keeps the earlier one in the history.
Go on. Get Harny.
The first 500 monthly-active end users are free for good. Create an app key and make your first call in minutes.