Features
Context follows the work.
A user should not have to re-explain their situation at every exchange. The agent keeps what helps the work move forward — and only that.
Four different things, often conflated
"Memory" covers distinct realities: what the application already knows, facts specific to your organisation, one person's preferences, and the state of the current task. Mixing them produces behaviour nobody can follow.
InApp-Agent keeps them apart. A preference expressed once in a specific context does not become a general rule, and a team instruction does not apply where it has no place.
What stays under your control
Visible provenance
Every retained piece of information carries its origin and scope. You can tell why the agent knows it.
Correction and deletion
An authorised user can inspect, correct, restrict or delete what has been retained.
Forgetting that holds
A deleted preference must not return through a summary, a cache or a restore. Pending work that depended on it is re-evaluated.
No infinite memory promised
What is kept serves a purpose. We do not promise to retain everything indefinitely: retention is defined with you.
Memory and training are distinct. The pilot defines what is retained, who can access it and for how long.
Frequently asked questions
Is our data used to train a model?
Memory retains permitted information to resume your work; it is not used to train the model, and your data is never used for other customers. Model-provider retention and data-use terms are checked for the chosen service and configuration, then specified in your proposal.
What happens if the session is interrupted?
Facts and choices explicitly remembered are stored before being confirmed to the user. A disconnection must not erase a promise already made.