Can a background job bring deleted AI agent memory back?

A deletion can appear to work while a queued write still holds a copy of the data. When that job completes, the information may become retrievable again.

We built Memtru to help developers reproduce memory isolation and lifecycle failures, then rerun the same tests after a fix.

The public integration demo follows a synthetic marker across an HTTP API and a separate background worker:

1. Write the marker and confirm it can be retrieved.

2. Hold an update in flight.

3. Delete the memory and verify it is absent.

4. Release the pending update and check again.

The deliberately vulnerable version returns the deleted marker and fails. The corrected version keeps it unavailable through the tested lifecycle events and passes. This is a synthetic sandbox example, not a customer deployment or a claim of physical erasure.

Try the demo: https://memtru.com/demo/integration

Run the local CLI demo: pip install memtru, then memtru demo. Testing your own system requires an adapter and an authorized sandbox.

For developers working with persistent agent memory: how do you currently test deletion when queues, retries or background tasks are involved? Which part is hardest to reproduce?