Goal
What the team is trying to make true
Persistent context · AI agent memory
AI agents that remember the work
Carry the relevant goals, decisions, evidence, and project history into the next task—so ongoing work can continue without starting from an empty prompt.
Continue the work without reconstructing it
The useful unit of memory is not every token ever seen. It is the relevant work record the team can recognize and the agent is allowed to use.
Decisions
What was approved, changed, or ruled out
Evidence
The files and results that support the work
History
What already happened and what remains
Before and after
The next run starts with the last decision attached
A team should be able to resume from the project state—not paste yesterday’s conversation back into a new window.
Without persistent context
“Can you remind me what we decided, which file won, and what remains?”
With Offloop work memory
Approved direction, evidence, owner, and next step retrieved from the same work record.
Capture the record. Retrieve what matters. Continue.
Memory should remain attached to a clear workspace and task boundary instead of becoming an invisible pool of unqualified recall.
Capture
Keep goals, progress, files, and decisions in the ongoing work record.
Retrieve
Bring relevant, authorized context into the task that needs it.
Continue
Resume with the prior outcome and next step visible to the team.
Memory boundary
Persistent context is not unlimited recall.
ScopedMemory belongs to the work and workspace that created it.
RelevantRetrieve what supports the current task, not every prior message.
Permission-awareLosing access to the source must not create new authority.
InspectableThe work record remains visible to the people responsible for it.
Questions about agent memory
What memory means, what persists, and what it does not imply.
What is AI agent memory?
AI agent memory is the product layer that helps an agent retrieve relevant prior context—such as goals, decisions, files, and work history—when it continues a task.
How is agent memory different from a long context window?
A context window is the information available to a model during one inference. Persistent memory is selected product context that can be retrieved again across later work.
What can an Offloop agent remember?
Offloop can preserve the work record around Channels and Runs: the goal, visible progress, files, decisions, and outcomes that belong to that work. Exact availability still follows product permissions and scope.
Does agent memory mean Offloop trains on my data?
No. Product memory and model training are different concepts. This page describes retrieval of authorized work context, not a claim that the model learns permanently from every interaction.
Can an agent remember everything forever?
No. Memory should be relevant, scoped, and permission-aware—not unlimited recall. Access and retention follow the workspace and product controls that own the underlying work record.
Keep moving