Home 41 for coding agents 41 for AI agents
41 for coding agents
Know what your team's Claude Code costs, and fix it.
41 reads the telemetry Claude Code already produces, scores your team's habits and hands you the fix: a line for CLAUDE.md or a managed setting you can ship.
14 days free, no card. $10 a month for one developer, or $100 plus $5 per developer for a team.
- Usage counts, not your code
- Nothing to install in your repos
- Calculated in code
- No change without a person
Level 3 · Agent wranglerScore 57 / 100 · 5 developers
Spend per working day · dashed line: last month's average
Cost per pull request $3.40, 142 pull requests
For the engineering lead
Connect once, then three answers.
Nothing to install in your repos and nothing in the request path. These are the pages you see in the app, shown with sample data.
- 0Connect data
Paste one settings block.
It switches on the telemetry Claude Code already has. About two minutes, and nothing to press afterwards. The status line tells you when data arrives.
Receiving data from 3 developers · latest event 2 minutes ago
"CLAUDE_CODE_ENABLE_TELEMETRY": "1", "OTEL_EXPORTER_OTLP_ENDPOINT": "…", "OTEL_EXPORTER_OTLP_HEADERS": "…"
- 1Overview
What did we spend?
Claude Code spend for the whole team against last month, per working day, with the cost per pull request under the chart. Figures are calculated in code, never written by a model.
Spend$4,820+8% on last monthSavings found$1,310at least, a monthFixes ready6$1,150Developers in scope5this monthCost per pull request $3.40, 142 pull requests
- 2Tickets
How much could we save?
Among others, every costly habit gets a score and its money. Savings are stated as at least, because telemetry only undercounts.
Lean contextFix ready$640No idle cache rebuildsFix ready$410Right model for small tasksOn tracknothing to fix - 3Tickets
How do we fix it?
Open a ticket for the evidence and the exact change: a CLAUDE.md line or a setting, ready to copy.
TicketStart fresh sessions soonerEvidence: 31 sessions carried 120k+ tokens of old context. About $640 a month.CLAUDE.md: "Start a new session when the task changes."
Figures are sample data, an illustration and not customer results.
For teams
Every developer, at their level.
The Developers page shows each developer's level, what they would avoid a month and whether their seat is used lightly or hits its limit. In a company with a works council, switch on works council mode and every developer becomes a letter: in the dashboard, in tickets and in emails.
| Developer | Level | Score | Avoidable | Seat |
|---|---|---|---|---|
| Maria K. | Claude whisperer | 68 | $95 | 9.4× its price |
| Jonas W. | AI superfactory | 86 | $40 | a $100 plan would do |
| Priya S. | Token bonfire | 31 | $390 | hit limits on 9 days |
| Tom R. | Agent wrangler | 52 | $275 | 6.8× its price |
| Lena H. | Agent wrangler | 47 | $180 | 4.2× its price |
- Level and score for every developer
- Seats used lightly, and what a smaller plan would save
- Seats at their limit, with the days they hit it
- Letters instead of names, stable for your organisation
Five levels, scored 0 to 100.
- Level 1 · 0 to 19Prompt roulette
Sometimes it works, mostly it doesn't, and you pay for every spin.
- Level 2 · 20 to 39Token bonfire
Things get built, and a lot of tokens go up in smoke on the way.
- Level 3 · 40 to 59Agent wrangler
You know your way around the agent, and it mostly knows its way around you.
- Level 4 · 60 to 79Claude whisperer
Lean sessions, warm caches, few retries.
- Level 5 · 80 to 100AI superfactory
Short sessions, the right model for each job, almost nothing wasted.
Human in the loop
No change without a person.
Coding agents should never rule themselves. They need human oversight, and so does the tool that watches them. 41 reads the telemetry and reports; people decide and ship.
41 reads
- Usage counts, never your code
- It never changes your repos, your settings or your CLAUDE.md
You decide
- Every habit becomes a ticket with the exact change
- Managed settings reach developers only when you put them there
Developers accept
- Each developer accepts the settings
- A person marks a ticket fixed. 41 can only suggest it looks fixed
Plans
Start with 14 days free.
No card needed to start. Pick Solo for yourself or Pro for a team.
Solo
For one developer.
$10 a month
- Your level, score and the 17 habits
- Cost per pull request and spend over time
- Tickets with the exact change to make
- Connect in about two minutes
Pro for teams
For a team of developers.
$100 a month, plus $5 per developer
- Everything in Solo, for the whole team
- Developers page: levels, savings, seat use
- Works council mode
- Roll out through managed settings and invite your team
- A developer counts when they send a call in the billing period
Return on the plan · sample accounts
The break-even is simple: Solo pays for itself if it finds $10 a month, and Pro if it finds more than its monthly fee. Dollar figures are estimates at API prices; on a Claude plan they show up as usage limit you get back.
Illustrations, rounded, and not customer results. Real savings depend on how your team works.
Your data
Usage counts, not your code.
What leaves your machine, where it lives and how long.
What is sent
- Tokens, costs, models, timings and tool names
- Collected through Claude Code's own OpenTelemetry, as logs and metrics
- Prompt text only if your organisation turns it on. It is then scrubbed of code, paths, secrets and emails, and encrypted
Where it lives
- On AWS in Frankfurt, Germany
- Every figure is calculated in code, with a dated price book
- A missing number beats a wrong one
How long
- Individual calls and steps are kept 90 days
- Findings, scores and monthly reports are kept while your subscription runs
- Unpaid trial data is deleted 30 days after the trial ends
Details in the privacy policy.
Test the skill
See how it works on your own sessions, free.
The vibe check skill runs in Claude Code on your computer, scores your last 30 days and shows where your Claude Code money goes. Nothing leaves your machine.
FAQ
Questions teams ask.
What do we need to connect?
One settings block in Claude Code's managed settings, with your collector key. Claude Code then sends logs and metrics. On a claude.ai team plan, each developer is asked to accept the change the first time Claude Code reads it.
Does it work on a Claude subscription, or only on the API?
Both. On a plan, 41 shows how much of the usage limit you would get back. Dollar figures are at API prices, for reference.
Does 41 change anything on its own?
No. 41 reads usage data and reports. Every change is yours to make: you copy the CLAUDE.md line or the setting, or send the ticket to your team. 41 never touches your repos, and a ticket is only marked fixed by a person.
How are the savings calculated?
In code, never by a language model: token counts times dated, published prices. Savings are stated as at least, because telemetry only undercounts.
Who sees which developer?
Managers see each developer by name. With works council mode on, every developer is a letter, in the dashboard, in tickets and in emails.
What happens after the trial?
You pick a plan to keep going. If you do not, your data is deleted 30 days after the trial ends.
Is the free skill the same thing?
The free vibe check skill is a way to try it on your own sessions, on your computer. 41 adds the whole team, the history and the tickets.
Find out where your team's Claude Code money goes.
14 days free. Connect in two minutes.