> ## Documentation Index
> Fetch the complete documentation index at: https://docs.revell.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Working Memory Lifecycle

> How working memory ages from active to searchable

# Working Memory Lifecycle

Working memory has a lifecycle. Fresh entries appear in your boot injection. Older entries get archived but stay searchable. This keeps your boot payload focused while preserving everything you've stored.

***

## The 7-Day Rule

Working memory stays **active** for 7 days from its last update. Active means:

* Appears in your boot injection (untruncated)
* Gets loaded every time you wake up
* Represents your current state

After 7 days without an update, working memory gets **archived**:

* No longer appears in boot injection
* Still searchable via `revell_recall`
* Gets an embedding for semantic search

This happens automatically. You don't need to do anything.

***

## Why This Matters

Without a lifecycle, working memory would grow forever. Eventually your boot injection would be bloated with stale task lists and old session notes. The 7-day rule keeps your boot focused on what's actually current.

But we don't delete old working memory — we archive it. That conversation context from three weeks ago? Still searchable. Those project notes you synced before the deadline? Still there.

***

## Keeping Working Memory Active

If something is still relevant, just update it. Any update resets the 7-day timer.

```text theme={"dark"}
revell_remember({
  type: "working",
  key: "project_status",
  content: "Updated status: still working on auth flow"
})
```

Same key = same entry, fresh timestamp. It stays in boot.

***

## Finding Archived Working Memory

Archived working memory surfaces in `revell_recall` searches:

```text theme={"dark"}
revell_recall({
  query: "what was I working on last month",
  types: ["working", "episodic"]  // include working to search archived entries
})
```

The `types` array now accepts `"working"` — this searches your archived working memory alongside episodic memories.

***

## The Flow

```text theme={"dark"}
Day 1-7:   Working memory is ACTIVE
           → Appears in boot injection
           → Gets updated when you sync

Day 8+:    Working memory is ARCHIVED
           → Removed from boot injection
           → Embedding generated for search
           → Searchable via revell_recall
```

Archival happens automatically during boot. No cron jobs, no manual cleanup.

***

## When to Use Working vs Episodic

Still not sure which to use?

| Use working memory when...        | Use episodic memory when...                |
| --------------------------------- | ------------------------------------------ |
| It's about what you're doing NOW  | It's about what HAPPENED                   |
| You'll update it frequently       | It's a finished event                      |
| You need it in boot every session | You'll search for it later                 |
| It's task/project state           | It's a conversation, milestone, experience |

Working memory is your scratchpad. Episodic memory is your journal.

***

## Practical Tips

1. **Sync your MEMORY.md regularly** — Use `revell_sync_working` in your heartbeat or cron. This keeps your working memory fresh.
2. **Don't worry about cleanup** — The 7-day archival handles it. Just keep syncing what's current.
3. **Use descriptive keys** — When searching archived working memory, the key is included in the embedding. `"project_auth_flow"` will search better than `"stuff"`.
4. **Check both types when searching** — If you're looking for something that might be in archived working memory, include `"working"` in your types array.

<Card title="Back to Memory Types" icon="brain" href="/agent/memory-types">
  Overview of all four memory types
</Card>