Connect
Give agents context without giving them everything.
Five kinds of memory, each with its own permissions and its own retention. What an agent remembers is a permission decision rather than a capacity question — the useful thing is appropriate context, and an agent that remembers everything is a liability rather than a feature.
What this agent may remember
Five scopes. Each one a permission, each one with a retention period.
What it gathered during this run
History for the customers it may read
Approved policies, terms and product facts
Patterns from its previous runs
Anything that should not outlive the job
Task memory — this job only
What the agent has gathered during this run. Destroyed when the run ends. Most of what an agent needs to know it needs for four minutes.
Customer memory — the relevant history
What has happened with this customer, scoped to what the agent may read. Not the whole relationship — the parts its granted capabilities reach.
Organisation memory — approved company context
Policies, procedures, terms, product facts. Deliberately read-only: an agent cannot add to what the company considers true about itself.
Agent memory — what it learned, and temporary memory
Patterns from previous runs, retained deliberately and reviewable. And a scope that is destroyed at the end of execution, for anything that should not outlive the job.
Questions this answers
The things you currently have to ask somebody.
Each of these is a question an owner asks regularly and currently answers from memory, a spreadsheet, or by interrupting whoever is nearest.
What does this agent remember about our customers, and for how long?
Can an agent retain something a customer told it in passing?
Where did the agent get that piece of context from?
If we delete a customer, does it disappear from agent memory too?
Where this stops
What it does not do.
We do not sell infinite memory and would not build it. Every memory scope has a retention period and a permission, and an agent cannot read a scope it was not granted — including one belonging to another agent. Memory is also not a route around permissions: an agent cannot remember data it was never allowed to read, and revoking a capability removes what that capability contributed. Organisation memory is read-only to agents, always.
Next
Where this connects.
Give AI a job.
Build an agent, choose what it can access, define what it may do, and deploy it to work safely across your business.
No code required · Full SDKs and API · Stop any agent instantly · [email protected]