Home 41 for AI agents 41 for coding agents
41 for AI agents
Know what every AI agent costs, and what to change.
The bill arrives as one number. The causes sit inside thousands of runs. 41 reads the data your AI agents already produce and gives you a cost account you can sign off, with the fix attached.
- Read-only: no SDK, no proxy
- Every figure traced to the run
- Calculated in code
- No change without a person
Spend by agent
99.2% of calls priced · 0.8% not priced, shown separately
For the person who owns the AI budget
Three questions, answered in three pages.
From the bill to the fix, in the order you are asked. These are the pages you see in the app.
- 1 / 3Overview
What did we spend?
Your AI agents' spend for the day, week, month or year, against the period before, by agent and by model. Calls with no price are shown, never hidden.
$62,400this month · +12% on last - 2 / 3Agents
How much could we save?
The waste 41 finds in each agent's own calls, valued in dollars by arithmetic, from findings already stored.
$11,856could be saved · 19% of spend - 3 / 3Tickets
How do we fix it?
Every finding becomes a ticket an engineer can act on, with an effort estimate and a risk rating.
Openeffort S · risk low
Figures are sample data, an illustration and not customer results. With 41 for coding agents as well, a Summary shows both products in one place.
What it finds
Where the money goes, and the runs it came from.
Among others, 41 looks for these patterns. Pick one to see the calls behind it. Figures are calculated in code from token counts and a dated price book, never written by a model, and each check reports one of three states: found something, measured and found nothing, or could not measure.
- Only measured spend is shown, with no extrapolation
- A missing number beats a wrong one
- Calls with no price are shown, not hidden
Sample data, an illustration and not customer data. Among others: more detectors are added as new ways to save cost appear.
Your data
Works with what you already collect.
41 can only be as good as the data you can provide, preferably OpenTelemetry instrumentation. Under Connect data in the app there are eleven ways to send it, and we add collectors as we go.
You send it
- OpenTelemetryPush with a collector key
- File uploadAny export: 41 recognises the format
- Custom JSONAny other tool, mapped once
- Claude CodeFor 41 for coding agents
41 reads it, read-only
- LangfuseTraces per agent
- LangSmithRuns per agent
- HeliconeRequests per agent
- PhoenixTraces per agent
- DatadogLLM Observability traces
Checks the bill
- OpenAI usageAccount totals by model and day
- Anthropic usageAccount totals by model and day
These never count as spend. They show how much of the provider bill your traces explain.
Each connection is switched on together with you when you start, so a source shows as waiting until then. Nothing goes in your code or in front of your models: no SDK, no proxy. If 41 stopped tomorrow, nothing your AI agents do would change.
Human in the loop · for two readers
No change without a person.
AI agents should never rule themselves. They need human oversight, and so does the tool that watches them. 41 reads and reports; one person decides and another ships. 41 never writes back.
- Your AI agentsKeep running, untouched
- 41Reads the data, finds the waste, writes the ticketReads only
- The executive decidesNinety seconds on the Summary, then a number to say out loudA person
- The engineer shipsA ticket that stands on its own: evidence, a prompt for their coding agent, the savingA person
- Someone confirms41 says "Looks fixed" and offers to close the ticket. You click.A person
41 never opens pull requests, never touches your code and never changes your agents or prompts.
Pricing
3% of the agent spend 41 measures, or a minimum of $299.
The fee for a month is the greater of $299 and 3% of that month's measured spend. You pay little when you spend little, and the fee grows only with the bill 41 is there to reduce. Claude Code usage is not part of this fee.
$63,000
The illustration assumes savings of around 25% of spend. Real savings depend on your AI agents.
The saving is measured on your own data before anyone changes a line of code, so you can check the return before you act.
Examples, rounded. Only spend that was sent and priced is measured. The sample account is an illustration, not a customer result.
Design partners
Shape it with us.
We are onboarding design partners for first trials. Try 41 on your own AI agents early, and help us build the value you need.
FAQ
Questions we get.
What do we have to send you?
Whatever data you have about your AI agents, preferably instrumented telemetry such as OpenTelemetry. We can only be as good as the data you can provide, and we add collectors as we go.
Do you see our prompts, and where does the data sit?
41 works on derived facts such as token counts, costs and timings. We are starting with hosting in Frankfurt and can host elsewhere on AWS and GCP. Where data is held is set in the data processing agreement.
How is the savings number calculated?
In code, never by a language model, on your own data. Backed by research and current developments in the AI space; more detectors are added as new ways to save cost appear.
How is this different from the dashboards we already have?
Your dashboards tell engineers what happened. 41 draws the line from a design decision to the money it cost, and ends in a change your team can ship.
Which AI systems does it support?
Anything that emits data about your AI agents. 41 reads OpenTelemetry, file uploads and custom JSON, and connects to Langfuse, LangSmith, Helicone, Phoenix and Datadog, with OpenAI and Anthropic usage as a check on the bill. We can only be as good as the data you can provide, preferably OpenTelemetry instrumentation.
Does 41 change anything on its own?
No. 41 reads data about your AI agents and reports what it finds. Every change is made by a person on your team: someone sends the ticket, your engineers ship the fix, and someone confirms it is done. 41 never opens pull requests or touches your code.
How is it priced?
3% of the agent spend 41 measures, with a minimum of $299 a month per organisation. Slide the calculator above to your own spend to see the fee and the break-even. Claude Code usage is not part of that fee.
Can we use it today?
We are open for design partners for first trials. Get 41 now, or ask for a free test, and we will be in touch to set up a first trial.
Visibility is good. Action is better.
41 ends in a fix you can ship.
← Back to the main page · Test the free vibe check skill for Claude Code