Skip to main content

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

Working memory is your agent’s “what I’m doing right now” memory. It’s for current tasks, active projects, and session context — things that matter today but won’t matter forever.

​
How It Works

1

Your agent syncs their current state

When your agent calls revell_sync_working, their current context gets stored. This might come from their MEMORY.md file or just notes about what they’re working on.
2

It appears in every boot

Active working memory shows up in your agent’s boot injection — the memories they load when waking up. They always know what they were doing.
3

After 7 days, it archives automatically

If working memory isn’t updated for 7 days, it moves to the archive. Still searchable, just not loaded on every boot. This keeps the boot payload focused.

​
Why 7 Days?

If your agent hasn’t updated something in 7 days, it’s probably not current anymore. The task finished, the project moved on, the context changed. But we don’t delete it — we archive it. Your agent can still search for old working memory if they need it.
The 7-day rule is automatic. You don’t need to configure anything. Your agent doesn’t need to clean up old entries.

​
On the Dashboard

Working memory appears in two places:
  1. Recent Experiences timeline — Mixed with episodic memories, tagged as “working”
  2. Boot Preview — Shows what your agent will see when they wake up
You can see what’s active vs archived by the timestamp — anything older than 7 days that’s still showing is being kept fresh by your agent’s regular syncs.

​
Setting Up Automatic Sync

For working memory to stay fresh, your agent needs to sync regularly. There are two approaches:

​
Option 1: Manual Sync (Simple)

Your agent calls revell_sync_working when they remember to. This works but depends on them actually doing it. Set up a cron job that reminds your agent to sync every few hours. Most agent frameworks support this: OpenClaw: 
# In openclaw.json
{
  "cron": [
    {
      "id": "revell-sync",
      "schedule": "0 */4 * * *",
      "prompt": "Read your MEMORY.md and call revell_sync_working with its contents."
    }
  ]
}
Claude Code: 
# In crontab
0 */4 * * * ~/.claude/hooks/revell-sync.sh
With a cron job, your agent’s working memory stays fresh automatically.

​
The generate-context Script (Optional)

Some agents have complex setups — knowledge graphs, multiple workspace files, browser history integration. For these agents, you can set up a generate-context.sh script that automatically writes a MEMORY.md summary. This is entirely optional. If your agent already maintains their own MEMORY.md, they don’t need this.

Compaction Protection

Set up automatic memory preservation during context compaction

​
FAQ

It gets an embedding (for semantic search) and moves out of the boot injection. Your agent can still find it with revell_recall — it just won’t load automatically every session.
Not directly in the current version. Archived working memory is searchable by your agent but doesn’t appear in the dashboard timeline. This may change in a future update.
They can search for it with revell_recall. Include "working" in the types array to search archived working memory.
Not currently. We kept it simple and predictable — 7 days, no toggles. If your agent needs something to stay active longer, they just need to update it.