Come back to a warm session.
Ember keeps the Claude Code sessions you’ll come back to warm, and prepares a compacted copy of very large sessions before it’s time to let go.
macOS 14 or later · Apple silicon
A small app in your menu bar.
Your sessions, your limits.
A break shouldn’t mean a cold start.
Leave a long session for lunch, a meeting, or overnight, and its prompt cache can expire after 5 minutes or 1 hour, depending on the cache tier. Your first message back re-reads the whole context at the cache-write price. That can cost more usage and make the first reply slower.
A small request while the session is still warm can read the cached context and refresh its timer.
| What happens | Estimated cost |
|---|---|
| Return after it goes coldWrite the full context again | $1.60 |
| Keep it warm for one more hourOne cache read of the same context | $0.04 |
Estimates at API list prices: $4 per million input tokens, $0.20 per million cache-read tokens, and a 2× write rate for the one-hour tier. The tiny reply’s output cost is rounded away. Subscription users consume their plan’s usage; Ember does not bill you for these requests. Actual prices and cache reuse vary.
Keep warm. Have a copy ready.
See what’s still warm.
Ember watches your Claude desktop sessions. Countdown rings show how long each cache has left, so you can see which sessions need attention.
Keep the ones you’ll return to.
Tiny refreshes run inside your own sessions, just before the cache expires. Your budget and quiet hours set the limits; staggered timing (jitter) spreads refreshes out.
Continue in a smaller copy.
For very large sessions, above 300k tokens by default, Ember can prepare a compacted copy before letting the cache go. Choose Continue in copy when you return. Your original stays available.
The companion plugin ships inside the app. Onboarding installs it into Claude Code with your approval, so refreshes happen inside each session’s own process.
You decide how much to keep warm.
Start with Smart, or choose a strategy that fits how you work.
- Smart Default
- Keep sessions warm, then prepare copies of very large sessions when it’s time to let go.
- Keep warm only
- Refresh within your limits. Don’t prepare copies automatically.
- Copies only
- Prepare compacted copies of very large sessions, without keep-warm refreshes.
- Watch only
- Show session status without issuing refresh or summary requests.
$2/day default budget. Smart stops refreshing when the budget is reached. Copies also use Claude usage; the budget is a policy limit, not a billing guarantee.
Quiet hours and a warm window. Defaults are 20:00–09:00 and up to 3 hours after your last message. Both are adjustable.
Pause whenever you need. For an hour, until tomorrow, or until you turn Ember back on. Set a session to Never touch, or choose Let it go to stop refreshing it until your next message.
Things to know
- Ember works with the Code tab in the Claude desktop app. Refreshing needs a running session with the companion connected.
- Refreshes and summaries are ordinary small requests made inside your own sessions. They use your plan’s usage or API credits.
- Ember is independent, and is not affiliated with or endorsed by Anthropic. Automated requests may be restricted by the terms of your plan; check them if that matters to you.
- Ember relies on undocumented Claude desktop internals. It self-tests and falls back to Watch only when a critical compatibility check fails.
- A sleeping Mac cannot refresh a cache. Closing the lid can also leave too little time to prepare a copy.
Your sessions stay yours.
Ember’s records, summaries, and cost ledger stay on your Mac. No account, no analytics, no Ember session server. This website uses no cookies or tracking.
Refreshes and summaries send session context to Anthropic through Claude, in the same way your normal Claude requests do. Sparkle checks ember.eighteenhq.com/updates/appcast.xml and downloads app updates. Update checks don’t upload your session content.
Read the privacy detailsFrom download to menu bar.
Download Ember.
Download for macOS and open the DMG.
Drag it to Applications.
Open Ember from your Applications folder. Look for the ember in your menu bar.
Connect the companion.
Onboarding shows the install commands and asks before running them. Send a message in a Claude Code session to connect it, then choose your strategy, budget, and quiet hours.
A few questions
Does it work with terminal Claude Code sessions?
This version targets the Code tab in the Claude desktop app. Terminal sessions are outside its current scope.
Does Ember add another usage bill?
Ember’s refreshes and summaries use your existing Claude plan or API credits. The dollar figures on this page are API-price estimates, not charges from Ember. Requests can use plan limits, and savings depend on the size of your sessions, cache reuse, and when you return.
What happens if Claude stops a session?
Ember needs the companion running inside the session to refresh it. Claude can stop idle sessions under memory pressure. Ember shows those sessions as Not tracked; it cannot restart their processes. Send a message in Claude to reconnect a session.
Is a compacted copy the whole conversation?
The copy continues from a model-written summary, so it can lose detail. Its first turn creates a fresh, smaller cache. The original remains available, and you choose when to Continue in copy.
Can it keep a session warm overnight?
Only while your Mac is awake, the companion is connected, and your policy allows refreshes. Smart’s defaults include a three-hour warm window and quiet hours from 20:00 to 09:00. Adjust those limits to suit your work; a sleeping Mac cannot refresh.
Can I leave a session alone?
Yes. Choose Never touch for that session, Let it go to stop refreshing until your next message, or Pause Ember. Watch only shows status without refresh or summary requests.