Lifecycle Policies

Remem 0.5 uses named lifecycle policies rather than a fixed memory-type switch. The policy controls retention, decay, archiving, and forgetting while the memory remains an ordinary searchable record.

Creating a memory

{
  "content": "The team moved the service to eu-west-2.",
  "policy": "long_term",
  "ttl_seconds": 0,
  "importance": 0.8
}

Use a short-lived policy with a positive ttl_seconds for temporary context. Use a long-lived policy for preferences, decisions, and facts that should remain available until lifecycle rules or an explicit delete retire them.

What the lifecycle does

• Retention expires records that have passed their policy window.
• Decay lowers importance or health as a memory goes unreinforced.
• Recall records meaningful use and can restore a memory’s lifecycle budget.
• Promotion moves a memory to a more durable policy when the application decides it matters.
• Archiving removes a memory from active search while retaining it for history or recovery.
• Hard deletion removes the record and its derived indexes permanently.

Maintenance runs through durable per-tenant jobs. It continues after a restart, and bounded runs keep large corpora responsive while continuation jobs finish the remaining work.

Choosing a policy

Keep policy selection close to the meaning of the data:

Data	Typical policy
Temporary conversation detail	Short retention
User preference or stable fact	long_term
Sensitive or compliance-bound context	Explicit retention and audit settings
Protected high-value event	A policy with an appropriate protection window

Policies can be configured per tenant through the administrative API. Changing a policy affects memories as lifecycle maintenance next visits them, so operators should allow the maintenance queue to catch up after a change.