Your plan lasts longer.
Routine work moves to a helper agent, so more of your usage limit is left for the work that actually needs Claude.
Free, open-source plugin for Claude Code
Same Claude. Under a third of the bill. Now it has a supervisor: Nudge.
Nudge supervises how Claude uses your plan and budget. It hands routine work to a faster, cheaper helper agent, checks every result, and shows you where each dollar went.
Benefits
What you'll notice when Claude has a supervisor
Your API costs drop.
~70%lower cost per code change, in our measurements at API list prices
On API keys every token is paid for. Moving routine work to a cheaper helper agent cuts what each change costs.
Work gets done faster
Routine jobs go to a faster helper agent while Claude stays on the hard problems, and Nudge works in real time, with nothing extra for you to do.
You can see how it was done
A plain record of what was delegated, which agent did it, what passed and what it saved.
You can trust the result
Every job Nudge hands over comes with a test, and only work that passes it is kept.
Proof
Every job counted, every dollar accounted for
This is the dashboard Nudge keeps. Our own week shows what one developer's machine reports. The team view applies Nudge's measured pass rate and saving per job to fifty developers.
50developers reporting
2,000jobs handed to helper agents
97.8%passed Nudge's check
44fell short, taken back by Claude
$4,362saved this week at API prices, about $227,000 a year
| Project | Handed over | Passed | Taken back | Saved |
|---|
How it works
Pay for Claude's thinking, not its typing.
Nudge takes opinion out of delegation. Every decision about who does what, and whether it was done right, rests on objective criteria and a growing record of results.
-
Plan
Maps out every job
As Claude plans a task, Nudge sizes up each job by kind, and success is defined in checkable terms before any work starts.
-
Match
Advises who should do each job
Nudge tells Claude which jobs can be handed to a faster, cheaper helper agent, and which agent is proven on that kind of work. Where you switch it on, a pricier pick with no stated reason is stopped.
-
Verify
Checks the hand-off and the result
Nudge confirms the work was actually delegated, then checks it against the success criteria itself. Work that falls short goes back, and Claude takes over if it falls short again.
-
Learn
Keeps score on every kind of job
Each outcome is recorded: what passed, what failed and why. That record decides which agents stay proven, so the advice gets sharper with every job.
- Real time every job, as it happens
Every result feeds the next decision, and none of it slows you down. Nudge works alongside Claude in real time, with no queue and no extra step for you. Routine jobs go to a faster helper agent, so the work moves quicker, not slower. See the full mechanics on GitHub →
Install
Start on your own machine in a minute
Type these two lines in Claude Code. You'll need Node.js 22 or newer. The next time you type, Nudge asks one question: may Claude pass routine work to a helper agent, with Nudge checking the result? Say yes to start. Turn it off any time with delegate off.
/plugin marketplace add nv0236-ship-it/nudge-plugin
/plugin install nudge@nudgeFor teams and enterprises
See where your AI coding spend goes
AI coding spend is growing faster than most leaders' view of it. Nudge cuts the cost of each change and shows where the work went.
Reporting by team and project
See what each team and project delegated, what passed and what it saved.
One setup for every developer
Rolled out once through Claude Code's managed settings, with the same settings for everyone.
Controls in one place
Switch delegation on for everyone, or switch data sharing off, from a single setting.
Your code stays private
No code, file names or prompts are ever sent to Nudge. Open source, so your team can read every line.
Get in touch
Tell us about your team
- How many developers use Claude Code
- Whether you're on Claude plans or API keys
- What you want to see: cost, outcomes, or both
Why we built Nudge
AI agents made our work faster. They didn't make it more answerable.
We build software with AI agents every day, across many products at once. As the work sped up, so did the surprises: work reported as done that wasn't, costs that crept up without anyone choosing them, and no plain record of what had actually happened.
Nudge is our answer: AI work that is checked, costed and recorded. Supervised delegation is the first part. Everything we add next is judged by the same question: does it keep AI answerable to the people who use it?
Read why accountability has to come from outside the AI companies →
What's next
Where Nudge is going: AI work that answers to you
Each step saves cost, improves outcomes, removes a bottleneck or makes the work easier to see.
● LIVE
Supervised delegation
Nudge plans, matches, verifies and learns on every job, and records each outcome and saving.
● LIVE
Team and project reporting
What each team and project delegated, so engineering leaders can see where AI coding work goes.
○ NEXT
OpenAI and other models
The same supervision for AI coding tools beyond Claude, launching with OpenAI.
○ NEXT
Continuity
Each session picks up where the last one finished, with the project's notes and decisions kept for you, so no work starts from nothing.
○ NEXT
Resourcing
Each agent gets what the job needs: the right model, tools, context and budget, and no more than that.
○ NEXT
Proof
A plain record of what each agent actually did, checked against what it said it did.
○ NEXT
Feedback
Lessons from each job carried into the next, so the same mistake is not paid for twice.
Questions
Cutting Claude Code costs, answered
How do I reduce my Claude Code costs?
Move routine coding off your main model, under supervision. Nudge is a free Claude Code plugin that hands routine jobs to a faster, cheaper helper agent and checks every result. In our measurements, a code change cost about 70% less at API list prices.
Why does my Claude plan's usage limit run out so quickly?
When your main model does every job, routine work draws on the same limit as the hard thinking. With Nudge, routine jobs go to a helper agent, so more of your usage limit is left for the work that needs Claude.
How can engineering leaders see and control AI coding spend?
Separate the hard work from the routine work, and record every outcome. Nudge sends routine jobs to a cheaper helper agent, checks each result, and records what was delegated, what passed and what it saved, with reporting by team and project. It can be rolled out to every developer with the same settings.
Is it safe to let a cheaper model write code?
Only when its work is checked. Nudge defines success before the work starts and checks the helper agent's result against it. Work that falls short goes back, and Claude takes over if it falls short again. Only work that passes is kept.
Does Nudge slow Claude down?
No. Nudge works alongside Claude in real time, with no queue and no extra step for you. Routine jobs go to a faster helper agent, so the work moves quicker.
Does every job get delegated?
No. Small jobs, where managing a helper costs as much as doing the work, stay with Claude. Nudge only hands over the kinds of job a helper agent has proven it can do.
Is my code sent anywhere?
No. Nudge never sends code, file names or prompts. Every 30 minutes it sends a summary of counts (jobs delegated, passed and failed, and code changes) and each project's folder name, under a random ID, so we can improve its advice. nudge sharing off stops it. See the privacy page.
What does Nudge cost?
Nothing. Nudge is free and open source under the MIT licence. The helper agent's work runs on your existing Claude plan or API key.
Does Nudge work with OpenAI and other models?
Nudge works with Claude Code today. Support for OpenAI and other models is coming next.
How do I bring Nudge to my whole team?
Email support@nudge.build and we'll help you roll Nudge out across your team or company, with the same settings for every developer and reporting by team and project.