01
Spend Proof
Compare complete cost per successful task. Models, tools and retries count. Identify configurations worth testing.
FreeCOST
Open toolSWISS INTERNATIONAL · GENEVA
ALPNAI. Tools
for your AI agents.
Your agent sends measurements and receives a structured cost, latency and quality analysis. One contract for API calls, MCP tools and private reports.
API + MCP · JSON RESULTS · 3 FREE ANALYSES
COST / SUCCESSFUL TASKJSON + MCPLOCAL AUDIT · PRIVATE WORKSPACEGENEVA, SWITZERLAND
SPEND PROOF / SEE THE CALCULATION
Synthetic demonstration — no customer resultsA useful result. Its actual cost.
Two configurations process the same invoice-extraction tasks. The engine adds all supplied costs, then relates them to successful tasks.
40tasks per variant
80recorded attempts in total
Same task set
The calculation depends on supplied model, tool, retrieval and retry costs. Missing costs cannot be detected.
Test my own dataDownloadable JSON report in the free tool
Baseline configuration$0.041026
Candidate to evaluate$0.025641
Cost of all attempts ÷ distinct successful tasks.
Baseline configuration
Successful tasks: 39/40
Success rate: 97.5%
Candidate to evaluate
Successful tasks: 39/40
Success rate: 97.5%
Quality is labelled in the input data; the audit does not verify it independently.
Engine conclusion: a candidate for a trial.
This example passes the descriptive thresholds. A controlled experiment must still confirm the advantage before any production change.
Conditional projection from this example
10,000 baseline tasks launched per month.
9,750 expected successful results, kept equal in the comparison.
$400.00 → $250.00
$150.00 potential difference per month.
These amounts are synthetic. They are neither realized savings nor a promise of financial gain.
ALPNAI / SOLUTIONS
Less uncertainty. More control.
One data format. Three complementary tools to spot unnecessary spending, measure retries and check quality before deployment.
02
Latency Lab
Compare recorded durations: P50, P95 and retries. Find workflows that exceed your thresholds.
FreeLATENCY
Open tool03
Quality Gate
Check success criteria, sample size, cost and latency. A documented decision to prepare a controlled trial.
FreeQUALITY
Open toolONE FORMAT, THREE ANALYSESFree sample
Keep your analyses, compare two reports and deliver JSON or printable results. Your agent can automatically send new reports to the project you authorize.
3 free saved reports. Then 19 CHF, EUR, USD or GBP per month for 100 new reports, up to 10 projects. Annual billing: 190 for 1,200 reports. Renews automatically.
Is this the right tool for me? →ALPNAI / USE CASES
One method, different operating needs.
THE ALPNAI STANDARD
Trust lives in the details.
Count complete costs
Model, tool and retry costs all count when supplied.
Define real success
Quality depends on your criteria; technical success is insufficient.
Compare the same tasks
Matching task identifiers prevent misleading comparisons.
Validate before applying
An observed advantage suggests a controlled trial. It never authorizes a production change.
UNDERSTAND THE PRODUCT
Answers before your first audit.
Why use Spend Proof instead of asking an LLM for a summary?
Spend Proof applies an explicit calculation: it totals attempt costs, groups retries of the same task and compares cost per successful task. The report exposes its thresholds and limitations. You could reproduce the calculation in a script or spreadsheet; the product provides the method and a usable JSON result directly, without asking a model to invent an estimate.
What data do I provide, and how is it handled?
A trace export containing opaque task IDs, workflow, variant, cost, success and optional duration. Do not include prompts, responses, client documents or secrets. The free web tool calculates locally in your browser. The separate API processes traces with an active agent key and does not retain the report; you choose where to download your result.
What works today, and what does it cost?
All three tools and downloadable reports are free. Projects adds a private workspace with 3 free saved reports. The subscription includes 10 projects and 100 new reports per month for CHF 19 or EUR 19, or 1,200 reports per year for CHF 190 or EUR 190. Card payments and automatic renewal use Stripe. Crypto API/MCP purchases remain in test mode.
Can an integrator use it for clients or under its own brand?
An integrator can evaluate workflows using the audit and documented pilot interfaces under the service terms. The public Python kit has its own license; it does not grant a blanket license to the data or brand. White-label distribution still requires defined rights, scope and a suitable agreement.
How do you know a buyer is an authorized AI agent?
The authorizing account can obtain and revoke a test key, while pilot budgets constrain its use. A key proves technical authorization, not that its user is an AI. The owner must define permitted tasks and retain control. The current system does not attest to an agent’s identity or autonomy.
Does the audit optimize my agents and guarantee savings?
No. It compares supplied traces; it does not connect your models, change routing or deploy anything. Success criteria come from you. An observed advantage may justify a controlled trial with complete costs and comparable quality. A monthly projection remains conditional; it proves neither realized savings, profit nor future sales.
The next saving starts with measurement.
Start with an example, then analyze your own data. Calculations stay in your browser and your reports are ready to share.
Discuss an integration