Open Datasets
5 datasets · 321 rows · version 2026.1· CC BY 4.0 · updated 22 August 2026
State of Fitness APIs 2026
25 rows (APIs, SDKs and models) · 7 columns · version 2026.1 · CC BY 4.0
The access-structure survey behind our annual report: for each product, how the vendor bills developers, what the end user must already own or pay for, and what approval stands between working code and a live product. One source link per row.
A coverage-driven sample, not a census: these are the products this site's own verified pages already cover in depth, so it implies nothing about market share. KinesteX, which funds this site, is one of the rows and is classified by the same rule as everyone else.
- Licence:
- Creative Commons Attribution 4.0 — reuse and redistribute freely, including commercially, with attribution.
- Cite as:
- AIFitnessAPI, "State of Fitness APIs 2026," version 2026.1, aifitnessapi.com, https://aifitnessapi.com/datasets/fitness-apis-2026.json
- Columns:
- id, label, category, devCost, userSideCost, approvalGate, source
Health Data Type Matrix 2026 (HealthKit ↔ Health Connect)
10 rows (health metrics) · 6 columns · version 2026.1 · CC BY 4.0
The cross-platform identifier mapping: each common health metric with its Apple HealthKit type identifier and its Android Health Connect record type, plus the gotcha where the two platforms quietly disagree — Apple stores HRV as SDNN, Health Connect stores RMSSD.
Scoped deliberately to the two on-device platform stores, because those are the cells we could verify against Apple's and Google's own documentation. A narrow table you can trust beats a wide one padded with guesses.
- Licence:
- Creative Commons Attribution 4.0 — reuse and redistribute freely, including commercially, with attribution.
- Cite as:
- AIFitnessAPI, "Health Data Type Matrix 2026 (HealthKit ↔ Health Connect)," version 2026.1, aifitnessapi.com, https://aifitnessapi.com/datasets/health-data-type-matrix-2026.json
- Columns:
- id, label, appleHealthKit, androidHealthConnect, watchOut, source
Fitness API Changes & Deadlines 2026
13 rows (dated ecosystem events) · 8 columns · version 2026.1 · CC BY 4.0
The dated record of what changed: deprecations, deadlines, term changes and model freezes. Every row is graded — confirmed means a vendor's own words are quoted on the linked page, reported means notices we could not confirm on an official page — and that grading survives into both formats.
Date precision is kept exactly as sourced — “2026-09” means a reported month, not a guessed day. The undated watch items on the tracker are excluded here because they carry no date to publish.
- Licence:
- Creative Commons Attribution 4.0 — reuse and redistribute freely, including commercially, with attribution.
- Cite as:
- AIFitnessAPI, "Fitness API Changes & Deadlines 2026," version 2026.1, aifitnessapi.com, https://aifitnessapi.com/datasets/fitness-api-changes-2026.json
- Columns:
- date, sortDate, title, summary, status, source, sourceLabel, verifiedOn
Fitness & Health API Glossary 2026
33 rows (defined terms) · 6 columns · version 2026.1 · CC BY 4.0
The controlled vocabulary: each term with its definition, the section it belongs to, the one page on this site that is its canonical explanation, and its stable identifier — so a machine reader can cite the exact definition it used.
These are the site's own editorial definitions, written against the vendor documentation cited on the linked pages — not quotations from a standards body. The groups are our sections, not an industry taxonomy.
- Licence:
- Creative Commons Attribution 4.0 — reuse and redistribute freely, including commercially, with attribution.
- Cite as:
- AIFitnessAPI, "Fitness & Health API Glossary 2026," version 2026.1, aifitnessapi.com, https://aifitnessapi.com/datasets/fitness-api-glossary-2026.json
- Columns:
- term, definition, group, href, anchor, id
HealthKit Type Identifiers 2026
240 rows (type identifiers) · 13 columns · version 2026.1 · CC BY 4.0
All four HealthKit naming families in one table — quantity, category, characteristic and workout activity — read straight from the JSON Apple serves to developer.apple.com, with units, value enums, platform versions, and the column Apple does not expose anywhere: whether a quantity type is cumulative or discrete.
Two columns are derived, not copied: aggregation and unit family. Apple states both in prose rather than as properties, so each was matched against Apple's own sentence and left null where Apple's wording does not say. Both apply only to quantity types — elsewhere null means the concept does not apply, not that Apple was silent. Apple's abstracts are reproduced to identify the API surface; the classification is ours, and Apple's documentation remains the authority.
- Licence:
- Creative Commons Attribution 4.0 — reuse and redistribute freely, including commercially, with attribution.
- Cite as:
- AIFitnessAPI, "HealthKit Type Identifiers 2026," version 2026.1, aifitnessapi.com, https://aifitnessapi.com/datasets/healthkit-type-identifiers-2026.json
- Columns:
- identifier, objcConstant, family, group, abstract, aggregation, unitFamily, valueEnum, iosIntroduced, watchosIntroduced, deprecated, appleDocumented, appleDocs
The rules these files follow
No dollar figures, by design. There is not one monetary value in any of these datasets. Most serious tiers in this market are contact-sales, published figures move within a quarter, and a stale number damages a citable source more than a missing one. What we publish instead is structure: how a vendor bills, what gates a launch, what the end user must already own. The cost planner works the same way and likewise refuses to produce a total.
A null means we could not verify it. It never means zero and it is never a guess. If a field is empty in the CSV, the corresponding page on this site carries no sentence documenting it — that is the whole claim. Judgement calls stay out of the data entirely rather than being laundered into a column.
Versioned, not overwritten. Every file is pinned at version 2026.1. The next edition re-verifies each row and ships under a new version rather than editing these in place, so a citation to 2026.1 keeps pointing at what 2026.1 actually said.
Every row traces to a page. Each dataset carries a link column pointing at the page on this site that states the claim in prose, with its own sourcing and review date. Nothing here asserts more than its source page does. How we verify is written up in our methodology.
Generated, not typed. Three of the four files are built from the same data modules that render the pages, so a dataset cannot drift from the page it came from. The counts, field lists and versions shown above are read out of the published JSON when this page is built, which is why they cannot disagree with the downloads.
Using them
The JSON files share one envelope: name, version, publisher, license, methodology, generatedFrom, a fields dictionary describing the columns, a stats block, and the items array. The CSVs carry the same rows and columns, RFC 4180 quoted, with an empty field wherever the JSON has null.
Attribution is the only condition. A link to aifitnessapi.com/datasets alongside the citation string above satisfies it. If you find an error in a row, the page it links is the place it will be corrected first.
Published by AIFitnessAPI. Independent and not sponsored — one row of the access-structure survey covers KinesteX, the company that funds this site, recorded on the same terms as every competitor.