Long Claude Code chat? Cut it in half before you resume.
/compact summarises and forgets. CCPrune goes through the whole chat and removes only what Claude no longer needs. Your decisions, constraints and open issues stay, so Claude picks up where it left off, carrying far less.
609k → 278k tokens
Free, and we intend to keep it free. If you feel it's worth something, choose your own amount.
Download CCPrune for MacFree, and we intend to keep it free. Choose your own amount if you want to contribute. Your license and the download link arrive by email.
macOS 14+ · Apple Silicon & Intel · Requires Claude Code
Smart, not just shorter.
Reads the whole chat.
Every part is judged on what it's still worth, not on how old it is.
Keeps what matters.
Decisions, constraints, edge cases and unfinished work stay in.
Drops the dead weight.
Repeats, stale output and finished detours go.
Your latest work stays exact.
And every prune can be undone.
Keep the context you worked for.
Your long chat holds the edge cases you caught, the approaches you ruled out and the constraints you explained. When those details are left out of a summary, you have to explain them again.
CCPrune removes redundant output and judges older work for what your next task still needs. Your latest work stays exact.
Decisions and constraints
A summary can lose early instructions and subtle details that matter later.
Older work is selectively kept, compressed or dropped.
Original detail
Older conversation is condensed. Exact wording may be absent from active context.
Selectively keeps original detail while shortening or removing finished work.
Recent work
Claude's partial-summary options let you keep recent messages intact.
The protected recent stretch stays word for word by default.
Cost of cleanup
Summarizing a large conversation is itself a large model request.
Local cleanup uses no model tokens. Haiku makes judgment calls only when needed.
Cost before you continue
Choose a focus for the summary. No side-by-side cost forecast.
Compare the next message and the next 100k, 300k or 500k tokens. See cold and warm cache estimates.
Recovery
The original transcript remains on disk. Claude Code also provides rewind.
A backup before every prune. Exact undo until you continue the chat.
Carry less history through every turn.
Every follow-up carries history back to the model. Removing unnecessary context can save input usage across the rest of your work. With a cold cache, the input reduction starts on the next request. With a warm cache, the smaller history can pay off over later turns after the initial cache rebuild.
See the trade before you prune. CCPrune forecasts both starting cache states and estimates the break-even point.
Illustrative forecast: 300k context → 80k, then another 100k tokens of work.
Includes $0.15 of estimated Haiku usage. Both paths assume the same future token growth and output.
Cold cache
$4.07 estimated
$2.07 estimated · 49% less
Warm cache
$2.64 estimated
$2.07 estimated · 22% less
Illustrative API-price estimates. Your forecast depends on your chat, model and continuation.
See forecast assumptions and subscription pricing
4k new tokens per turn, including 800 output tokens; demo rates of $4 input, $20 output, $0.20 cache read and $5 cache write per million tokens. Smart-pruning reduction and Haiku usage are estimates. The pruned chat starts with a new cache. Figures exclude any optional contribution. Subscription users see equivalent API costs; pruning reduces token usage rather than the subscription price. How history and compaction use tokens ↗
From a heavy chat to the next prompt.
-
Choose a chat.
See saved chats from terminal, desktop and IDE sessions.
-
Review the estimate.
Compare context size and indicative costs before pruning.
-
Stop, then prune.
Stop a running chat first. Recent work stays intact; optional Haiku makes decisions about older work.
-
Continue in Claude Code.
Resume the chat yourself. Exact undo is available only until you continue it.
What matters stays
Decisions, constraints and open issues stay in. Your latest work stays exact.
Estimates before you act
Compare the cost of continuing at different horizons.
Backup before each prune
Restore it byte for byte until you continue the chat.
$4.99 Free. Get your lifetime license key.
CCPrune is free, and we intend to keep it free. If you feel it's worth something, choose your own amount. Your free license unlocks unlimited pruning, including all future updates, on one Mac at a time.
Default $0 USD · Optional contribution · No subscription
A few things before you install
- macOS 14 or later, Apple Silicon or Intel
- Claude Code and Apple's Command Line Tools
- 5 free prunes on every Mac, no key needed. For unlimited pruning, your license key arrives by email from Dodo Payments.
CCPrune is not yet notarized by Apple; first launch may require Open Anyway in System Settings. See the installation guide.
Optional Haiku processing uses your Claude Code login. The free license does not include unlimited model usage.
Is it free?
Yes. Get a free lifetime license key for unlimited pruning. We intend to keep it free. You can choose your own amount if you want to contribute. You can also try 5 prunes without a key. Sizes, cost forecasts and undo stay free.
What if it doesn't help?
Email hello@under5dollars.com within 7 days of contributing and we'll refund your contribution in full. No forms.
Can I prune a chat that is still open?
You can see its size and estimated cost while it runs. To prune it, stop it first: a running Claude Code keeps the chat in memory. Resume the lighter saved chat yourself in Claude Code.
What does /compact forget?
Compaction can leave exact wording out of active context; the original transcript remains on disk. Claude Code also provides partial summaries and rewind. Smart pruning is designed to retain useful decisions and unresolved issues.
Both make judgment calls about active context; neither guarantees perfect recall. Anthropic documents that compaction can lose early instructions. How compaction works ↗
What if it cuts something I needed?
Every prune keeps a backup. Undo restores the saved chat byte for byte until you continue the chat. Recent work stays word for word; older work is selectively kept, compressed or dropped.
How accurate are the dollar amounts?
Token reductions and dollar amounts are indicative estimates. Dollar amounts use API list prices, not a bill or guaranteed savings. A pruned history starts uncached: if your current cache is warm, a short continuation can cost more after pruning. The app compares different horizons and shows the estimated break-even.
What if I use a Claude subscription?
You are not billed per token on a subscription. The figures show the estimated cost at API prices. Context and tokens still matter to plan usage, but no specific subscription saving is guaranteed.
Does anything leave my Mac?
The app has no telemetry. Optional smart pruning sends an outline of older work to Haiku through your own Claude Code login. Redeeming your key sends it to our server to verify your Dodo purchase and issue a lifetime license, never your chats. The redeemed license and backup are verified locally without periodic online checks. The license does not include unlimited model usage.
What do I need?
A Mac with macOS 14 or later, Claude Code and Apple's Command Line Tools, which include Python. If missing, the app explains how to install them. See the installation guide.
Why does macOS warn me the first time?
CCPrune is not yet notarized by Apple. The first launch may require System Settings → Privacy & Security → Open Anyway. See the first-launch guide.
Does my license expire?
Your license lasts for life and includes future CCPrune updates. Redeem the key within one year, then save your lifetime license backup. The redeemed license and backup keep working offline after the key expires. If all local copies are lost, you need your backup to restore access without contacting us.
Can I move it to another Mac?
Yes. Save your backup from CCPrune → License… → Save License Backup, then choose Restore License Backup on the new Mac. Remove the license from the old Mac if you still have it. Use it on one Mac at a time; restoring a backup needs no internet connection.
More from the same maker
Our other appsUsage HUD
Claude Code pluginPlan usage, context, cache and estimated cost above your prompt.
Buddy
Mac · Windows · freeYour own AI assistant on WhatsApp or Telegram, running on your computer with free AI models.
Jhilmil Stories
Android · in testingHindi bedtime stories with pictures, for parents to read aloud to children aged 3 to 6.