The commitments at a glance
- Uptime
- 99.9% per calendar month
- Covered
- This website and its API endpoints
- Measured on
- External probes, five-minute interval
- Data cadence
- One reading every 90 minutes
- Freshness target
- 99% of keyed calls within 180 minutes of a fetch
- Urgent support
- First reply within 4 business hours
- Normal support
- First reply within 2 business days
- Service credits
- Up to 50,000 rate and 5,000 conversion requests
- Version support
- 24 months past a successor's release
- Deprecation notice
- 12 months, by changelog and email
- Balance expiry
- None — credits never expire
Uptime commitment
Fiat Agent is a website that publishes currency reference data, and an API that serves the same figures. The target is 99.9% monthly availability for both — the pages a visitor opens and the endpoints a developer calls. In a 30-day month that allows about 43 minutes of unavailability.
Every page and every response is answered from the site's own stored reading, so availability is measured on the part of the chain this website controls: a request that returns a stored reading counts as available even if the upstream provider happens to be unreachable at that moment.
Live technical state — the reading being served, how old it is, and the last sync run — is published on the status page at every moment.
How it is measured
- Availability is measured from outside the hosting network: a probe calls the public endpoint every five minutes.
- A probe counts as available when it receives a 2xx response whose body parses and carries a rates object.
- Refusals the site is designed to give — 401 for an invalid key, 402 for an exhausted balance, 429 for a rate window, 400 for a malformed request — are excluded from both the numerator and the denominator. They are not downtime.
- Announced maintenance counts against the target only beyond the first 15 minutes of the window.
- The measurement behind a claim is shared with the account that makes it, so a credit is never decided on an unpublished number.
Service credits
There is no subscription here, so a missed commitment is settled in requests added to your balances. They do not expire, and they are added to both meters in the same ten-to-one proportion the site uses everywhere.
- Monthly availability between 99.0% and 99.9%: 10,000 rate requests and 1,000 conversion requests.
- Monthly availability below 99.0%: 50,000 rate requests and 5,000 conversion requests.
- A run of more than six hours in a month without a successful fetch is settled under the same credits, at the 99.0% tier.
- Claims are made within 30 days of the affected month, from the email address registered to the account, naming the endpoint and roughly when the failures happened.
- An approved claim is credited within 5 business days. Credits are the sole remedy for missed availability.
Data freshness
A reading is fetched from the upstream provider every 90 minutes and written to the store the whole site reads. The target is that 99% of keyed calls in a month are answered from a reading fetched within the previous 180 minutes.
- Every response states when the reading was fetched and, separately, when the provider published it. The rate pages show both.
- A reading older than the staleness threshold is flagged as stale rather than presented as current.
- The provider republishes on its own schedule, often once a day. A reading that has not changed since yesterday is reported as unchanged, with both timestamps, rather than relabelled as fresh.
- Changes between readings are calculated against the previous published reading, and the page or response says which reading that was.
Support and response times
Support is by email, answered by the person who writes and runs this website. Business hours are Monday to Friday, 09:00–18:00 IST (UTC+5:30); mail arriving outside them is answered on the next business day.
- Urgent — the site or the API is unavailable, or figures are wrong for every caller: first reply within 4 business hours, with work starting immediately.
- High — one account cannot sign in, be credited, or read its balance: first reply within 1 business day.
- Normal — questions, documentation and feature requests: first reply within 2 business days.
- Security reports: acknowledged within 3 business days, as stated on the security page.
Security commitments
What is not claimed: no SOC 2 or ISO 27001 certification is held, and none is implied by this page. The security page sets out the measures in detail, and the terms of service limit liability as they state.
- TLS on every page and every endpoint. A plain-HTTP request is never served.
- API keys exist only as SHA-256 digests, are compared in constant time, and stop working on the next request after revocation.
- Per-record access rules: an account's keys, balances and purchases are readable only by that account.
- Card data never reaches this website. Payments run on PayPal's hosted checkout, and an order is credited only after a server-side capture check.
- A confirmed incident affecting account data or keys is reported to affected account holders without undue delay, saying what was exposed and what to do.
- A data processing addendum for personal data can be provided on request.
Versioning and long-term support
The current API version is /api/v1. Long-term support means the version you integrate against keeps working, unchanged, while a successor is prepared beside it.
- Inside v1, fields are never renamed or removed. New fields may be added at any time, so clients should ignore what they do not read.
- A breaking change ships on a new path — /api/v2 — and v1 keeps working alongside it.
- When a successor is released, the previous version is supported for at least 24 months from that release.
- Retirement is announced at least 12 months ahead in the changelog and by email to every account holder, with the migration path documented.
- Keys, balances and purchase history carry over. A version change never resets what a key can do or what you have paid for.
Limits and quotas
Limits are enforced by the endpoints and published here rather than discovered in production. These are the same figures the responses report in their rate-limit headers.
- Unkeyed calls: 5 a second, 50 an hour and 50 a day.
- Keyed calls: 20 a second and 2,000 an hour per key.
- A rate-window refusal returns 429 with retry-after and the rate-limit headers. An exhausted balance returns 402 with the code quota_exhausted.
- An account opens with 2,000 rate requests and 200 conversion requests. Both meters are separate, and a refused call draws nothing from either.
- Nothing renews automatically and no balance expires. A top-up adds to both meters permanently.
- Limits may be raised or lowered; a reduction is published before it takes effect.
- History: up to 2,000 stored readings per request, with pagination for anything beyond that.
What does not change under you
- Readings are append-only. A sync writes a new snapshot and never rewrites an earlier one, so the history behind a figure cannot silently change.
- Nothing is worked out at the moment you look: every page and every response reads the same stored reading, so any figure can be traced back to the moment it was published.
- Every change to this website is listed in the changelog with its date. The figures on this page are read from the same settings the site runs with, so the page cannot promise a cadence the site does not keep.
- The pages and the API are published from the same code and the same stored reading, so a figure shown on a page and the same figure returned in a response are never different numbers.
What is not covered
- Refusals the site is designed to give: invalid keys, exhausted balances, rate-window limits and malformed requests.
- Failures in the visitor's or caller's own network, DNS, browser or client code.
- Announced maintenance inside its notice window, and the first 15 minutes of any maintenance window.
- An upstream provider publishing nothing new. The reading is then unchanged, and both timestamps say so; availability and freshness are measured on what this website serves.
- Events outside reasonable control, including a failure of the hosting platform itself.
Claiming a credit, and changes to this page
Write to duplexinterface@outlook.com from the address registered to the account, with the endpoint you call and the approximate time of the failures. You will get a reply with the measurement and, if the commitment was missed, the credit applied.
If these commitments change, the revised page is published here with a new date and account holders are told by email at least 14 days before the change takes effect. A commitment is never weakened to cover a period that has already passed.
Questions about this page?
Write from the email address registered to your account and we will answer directly.
duplexinterface@outlook.com