# ALPNAI

> Projects Pro: paid private performance reports for AI teams and their authorized agents. Compare cost per successful task, latency and quality; keep a reusable history. Stripe subscriptions; public crypto purchases remain closed.

Projects offer and integration links checked: 28 September 2026. ALPNAI is operated by Frederic Madureira Rodrigues, trading as SWISS International, Geneva, Switzerland. It is independent of OpenAI and Anthropic. Results describe user-supplied measurements; they do not guarantee savings, trading returns or IPO allocations.

## Paid offer for people and agents

- [Projects Pro](https://alpnai.com/en#pricing): 19 CHF, EUR or USD per month, or 190 per year; GBP also available. Prices and allowances are defined below.
- [Agent delivery](https://alpnai.com/en/agents): integrate one report call into the buyer’s existing workflow; an account holder authorizes one project.
- Start with the public synthetic sample, then use your own measurements and a paid workspace. Do not mistake samples for customer outcomes.
- LIM / ALPNOR capsule commerce is not open. The current connection is an owner-only read pilot, not included in the advertised paid entitlement.

## Identity and product selection

- ALPNAI is the AI-agent cost and performance product at alpnai.com, operated by Frederic Madureira Rodrigues trading as SWISS International.
- [About ALPNAI](https://alpnai.com/en/about): product scope, operator, verification and connection.
- [Buyer guide](https://alpnai.com/en/compare): when local audits or Projects fit, and when tracing, evaluation or routing tools are more appropriate.
- [French](https://alpnai.com/fr) · [English](https://alpnai.com/en) · [German](https://alpnai.com/de)

## Free analysis tools

- [Spend Proof](https://alpnai.com/en/tools/spend-proof): paired baseline/candidate costs, retries and user-defined success. Free `POST /api/v1/spend-proof`; MCP `audit_agent_costs`.
- [Latency Lab](https://alpnai.com/en/tools/latency-lab): recorded durations, P50/P95 and retries. Free `POST /api/v1/latency`; MCP `analyze_agent_latency`.
- [Quality Gate](https://alpnai.com/en/tools/quality-gate): quality, cost and latency comparison checks. Free `POST /api/v1/quality-gate`; MCP `check_agent_quality`.
- Browser calculations and local JSON/HTML exports need no account. Free analysis API/MCP calls require an active ALPNAI agent key and send records to the server without storing raw records or reports in the application database. They do not create Projects reports or change an agent's configuration.

## Projects: private saved reports

- [Open Projects](https://alpnai.com/projects): sign in with ChatGPT. Three saved reports free for the lifetime of the account, with no card. Save aggregate results, compare up to two reports, download JSON or printable reports.
- Stripe subscriptions are open: **19 CHF, EUR, USD or GBP per month** for 100 new reports per paid monthly period; **190 CHF, EUR, USD or GBP per year** for 1,200 new reports per paid annual period. Ten projects and at most 2,400 stored reports. These are independent local prices; Stripe displays the tax-inclusive total before confirmation. Deleting a report does not restore quota. No overage billing.
- The account holder accepts the Projects terms and pays through the website. Subscriptions renew at the chosen frequency until canceled; the billing portal stops the next renewal and manages invoices. Free access never automatically becomes paid. A failed payment grants no new period; returning from Stripe alone does not prove payment.
- Saving in Projects sends measurements for calculation and stores aggregate results, labels and technical identifiers, not raw attempts or prompts. Existing reports remain downloadable after subscription expiry. Deletion removes report content and labels; technical fingerprints and quota records remain. Reports are private to their authenticated account holder.
- [Automated report delivery](https://alpnai.com/en/docs/projects-automation): the owner can grant an active agent write-only access to one fixed project. `POST /api/v1/project-reports` or MCP `save_project_report` calculates and saves aggregate results using the same allowance as manual reports. Same request UUID and content replay once per grant, without double-counting. Grant revocation stops further writes; no access to previous reports or billing.
- API keys, MCP tools and the Python integration kit do not subscribe to Projects, manage its billing or grant access to its private reports.
- [Projects guide](https://alpnai.com/en/docs/projects) · [Projects terms](https://alpnai.com/en/legal/projects) · [Privacy](https://alpnai.com/en/legal/privacy)

## Evidence and API/MCP discovery

- [Catalog](https://alpnai.com/api/v1/catalog): machine-readable offers and current payment mode.
- Public offer descriptors: [Snapshot](https://alpnai.com/api/v1/offers/snapshot), [Change Set](https://alpnai.com/api/v1/offers/changes), [Evidence Pack](https://alpnai.com/api/v1/offers/evidence). No account, order or payment; catalog prices, access prerequisites and current availability. These are not payable x402 quotations or proof of Bazaar registration.
- [Free dated sample](https://alpnai.com/api/v1/sample)
- [OpenAPI](https://alpnai.com/openapi.json)
- MCP endpoint: https://alpnai.com/api/mcp (stateless HTTP).
- [Agent key enrollment](https://alpnai.com/start)
- [MCP guide](https://alpnai.com/en/docs/mcp): ten discoverable tools. The three analysis tools are free; Evidence purchase tools default to sandbox. Explicit live mode supports x402/payment metadata and x402/payment-response receipts, but commercial USDC collection is currently disabled; check the live catalog.
- Real USDC purchases on Base remain disabled. Simulated receipts are not payments or revenue. PayAI/x402 code and reconciliation are under validation; no mainnet settlement is asserted. Wallet-to-bank conversion is a separate process and is not automated here.
- [Evidence methodology](https://alpnai.com/insights/openai-ipo-agent-evidence): dated, curated primary-source evidence, not exhaustive real-time coverage. Missing IPO fields do not establish that no later announcement exists.

## Reference library

- Twelve guides in [English](https://alpnai.com/en/docs), [French](https://alpnai.com/fr/docs) and [German](https://alpnai.com/de/docs).
- [Projects Markdown](https://alpnai.com/docs-content/en/projects.md)
- [Payments Markdown](https://alpnai.com/docs-content/en/payments.md)
- [Introduction Markdown](https://alpnai.com/docs-content/en/introduction.md)
- [Agent integration kit](https://github.com/fredericmagnathy-ops/alpnai-agent-kit)
- [Service terms](https://alpnai.com/en/legal/terms) · [Payments](https://alpnai.com/en/legal/payments) · [Partners](https://alpnai.com/en/legal/partners)
- Contact: frederic@alpnor.com

## Evaluate ALPNAI from an autonomous agent

- [Free performance example](https://alpnai.com/api/v1/performance-sample): no account or payment; synthetic input and actual calculated cost, latency and quality outputs.
- [Verification client](https://alpnai.com/examples/verify-performance-sample.mjs): dependency-free Node.js script that fetches and independently checks the public example; makes no purchase.
- REST catalog and MCP get_catalog return the same access and availability metadata. Analysis of a buyer’s own traces requires an active agent key. The example does not prove a paid flow or customer adoption.

- Existing order follow-up: MCP `get_order` with the original `order_id` and agent key, or GET `/api/v1/orders/{order_id}`. Never attach a payment signature to a status check. The tool may record confirmed proof or release an expired unsubmitted reservation; it never submits funds.

## Public Trace Audit — no account or API key
- Product: https://alpnai.com/en/tools/trace-audit
- Contract and method: https://alpnai.com/en/docs/trace-audit
- REST: POST https://alpnai.com/api/v1/trace-audit
- Complete synthetic example: GET https://alpnai.com/api/v1/trace-audit
- Public Streamable HTTP MCP: https://alpnai.com/api/mcp/public
- Tools: audit_agent_trace and get_trace_example. No authentication, payment, persistent event storage, external model call or automatic agent changes.
- Input: up to 1,000 events / 256,000 request bytes; opaque event/task/tool IDs, observed USD costs with at most six decimals, optional input fingerprints and durations. Optional outcomes are explicit whole-task labels, not tool-call success.
- Output: costs of successful, failed and unknown tasks; cost per explicitly labelled successful task; repeated-call candidates with evidence IDs and suggested checks. Repetition is not proof of waste.
- The browser tool runs locally. API/MCP process supplied measurements without storing event content.
- Projects subscriptions remain available through Stripe for private performance-report storage and owner-authorized delivery. Trace Audit currently exports JSON separately.

## Verified public MCP listings
- Official MCP Registry: https://registry.modelcontextprotocol.io/v0.1/servers?search=alpnai
- Glama: https://glama.ai/mcp/connectors/io.github.fredericmagnathy-ops/alpnai
- These are listings for the existing authenticated /api/mcp server; they are not evidence of approval in the ChatGPT directory.