Same Claude plan. Up to 3× the work.
A Mac menu-bar app for Claude Code and Codex. It shows your limits and where every token goes. On Claude Code it also cuts the waste, so your limits last longer.
$100Claude Max 5x$24Overplusup to $300of Claude work
See your own numbers first. Pay only to turn savings on.
7-day money-back, no questions. macOS 14 or later.
- Works with
- Claude Code
- Codex
One real week. Most of it was re-reads.
92.5% of the bill was Claude re-reading and re-saving the chat. Replies were 5.5%.
$693at API list price
- Cache reads$433 · 62.5%Claude re-reading the chat so far
- Cache writes$208 · 30%The chat saved again after it changed
- Replies$38 · 5.5%What Claude actually wrote
- Thinking$14 · 2%Reasoning tokens
Overplus cuts that waste. Up to 3× the work on the same plan.
The builder's own Claude Code week, Sep 17 to 23, 2026, at API list price. Real calls, real bill. The cache worked. The chat was simply long.
Now check your own week.
Claude Code transcripts. Runs in this tab, nothing is uploaded.
Open this page on your Mac to check your own week.
- Open your Claude folder
- Drag the
projectsfolder here, or
The builder's Claude Code week, at API list price.
$693
Choose your projects folder to see your own before and after.
Does it make Claude dumber?
Hard calls stay on Opus. In the builder's week, the router kept 2,416 of 3,864 calls on Opus (estimated).
Stays on Opus 5.5
2,416 calls · hard · estimated- Debug the race condition in the job queue
- Refactor the auth module across the codebase
- Security audit of the upload handler
Moves to Sonnet 5
1,448 calls · estimated- Rename this variable to userId
- Format this file
- Fix the typo in the README
Safe to try. Easy to undo.
On Claude Code, each call makes one stop on your Mac, then goes to Anthropic.
Claude Code
your terminal
Overplus
127.0.0.1:8788
- Edits 3 fields on each call: model, effort, reply cap.
- Other changes are Claude Code settings you switch on.
api.anthropic.com
over HTTPS
Never sent to www.overplus.dev
Prompts, replies, code, file paths, repo names, MCP names, credentials.
Sent to www.overplus.dev
Email, license, device name, usage counts. Required. Privacy policy
- Switch a feature off and your old setting comes back. One Terminal command removes Overplus.
- If Overplus crashes, it restarts in seconds. If you quit it, or it keeps crashing, new Claude Code sessions go straight to Anthropic again.
- Closed source. Runs only on your Mac.
- Overplus runs on your own subscription. It changes only the model, effort and caps of your calls, and the Claude Code settings you switch on. We know of no account flagged for this, but we cannot promise Anthropic's or OpenAI's policy.
Optional Google Analytics is off until you accept on the website or enable it in the Mac app. It never receives prompts or code. Manage it in the privacy settings.
Cheaper than the next plan up.
Every $1 on Overplus adds up to $8 of Claude work.
Up to 3× the work for $5 more. The next plan up costs $80 more.
1×
Pro
$20
up to 3×
Pro + Overplus
$25
5×
Max 5x
$100next plan up
Height is the work each option covers, compared with Pro alone. Claude prices are per month.
Already a customer? Add a tool from your billing page.
Buy LightBilled today. 7-day money-back, no questions.
The app is free to download and measure. Paying turns the savings on.
Overplus Light for Claude Code, sized for Claude Pro or ChatGPT Plus: $5 a month.
One license: one user on one Mac, for the tools you pick.
Same app on every plan. The price follows the AI plan you use.
Up to 3× is for Claude Code. On Codex, Overplus sets the model and effort, and claims no saving. Your own week can show a different number. Check your own.
What it changes. What it sends. How to undo it.
On Claude Code, Overplus can change the model, effort and reply cap of each call. Prompts and code pass as sent. On Codex, it can set your default model, reasoning effort and automatic compaction. With a ChatGPT login, it can also route each Codex call through a local proxy on 127.0.0.1 and set the model and effort of that call. Prompts and code pass as sent there too. Codex command-line flags and trusted project settings can override the defaults.
Check it yourselfSee what the proxy seesOn Claude Code, up to 3× the work on the same plan. Your own week can show a different number. The app shows it from your own calls. The 3× figure is for Claude Code only. On Codex, Overplus shows your limits, spend and where tokens go, and can set the model and effort. We do not claim a Codex saving.
Check it yourselfCheck your own weekThe app is free to download and use in measure mode. It shows your limits, your spend, where your tokens go, and what each lever would save, as an estimate. Paying turns the levers on for the tools you pick: Claude Code, Codex, or both. The saving claim is for Claude Code only. 7-day money-back, no questions: email pranav0123.pm@gmail.com.
Check it yourselfSee pricingNo. Prompts and code go only to the tool's own vendor: Anthropic for Claude Code, OpenAI for Codex. That is where they go without Overplus too. We get your email, license, device name and usage counts, never your text.
Check it yourselfSee what we getSwitch a feature off and your old setting comes back. To remove Overplus, run /Applications/Overplus.app/Contents/MacOS/Overplus --uninstall in Terminal. Full steps: /uninstall
Check it yourselfEvery way to uninstallSpend your limit on work, not re-reads.
Quit Overplus, and new Claude Code sessions go straight to Anthropic.
See your own numbers first. Pay only to turn savings on. 7-day money-back, no questions.