Memory that expires on purpose

Cost, safety and controlTeam whose room got noisy

Six months in, old one-off answers crowded out useful ones.

How do I stop old one-off answers from crowding out useful AI agent memory?

Set expiry settings in ArcRoom with arcroom rule, for example arcroom rule --tag pricing --idle-days 14. Memory can expire after n hits or after d idle days, per tag, so answers about fast-moving things like prices and versions go quickly while decisions live longer. Expiry settings are per team and need someone to set them.

Also asked as: Our shared agent memory got noisy after six months. How do we clean it up? · How do I make AI agent memories expire after a set time? · Can agent memory forget answers about prices and versions faster than decisions?

The work

arcroom rule sets when memory expires: after n hits, or after d idle days, per tag.

ArcRoom keeps
  • Decision
  • Expiry rule
The next agent gets

$ arcroom rule --tag pricing --idle-days 14

Step by step

  1. arcroom rule sets when memory expires: after n hits, or after d idle days, per tag.
  2. Answers about fast-moving things (prices, versions) expire quickly.
  3. Decisions can live longer.

What the agent sees

$ arcroom rule --tag pricing --idle-days 14
rule set: {"tag":"pricing", … "idle_days":14}

Compared with today's tools

Manual pruning of CLAUDE.md or a Mem0 store

Someone reviews the memory file or store every few weeks and deletes outdated entries.

Setup
none
Ongoing
review session, ~30 min a month

ArcRoom

Every agent reads the team playbook before it works and writes to it when it finishes. Nothing to update by hand.

Setup
1 min per setting (arcroom rule)
Ongoing
None

A human review catches subtle staleness that rules miss; arcroom rule does the routine expiry without anyone remembering to.

Where it stops

Expiry settings are per team and need someone to set them.

Try it in your project

Works with the agents your team already uses, and any agent over MCP or the HTTP API.

More in cost, safety and control