generated: '2026-08-17' method: searched source: https://developers.kardinal.ai/ versioning: scheme: uri-path current: v2 spec_version: 2.55.0 docs: https://developers.kardinal.ai/guides/migrating-api-versions policy_published: false note: >- The migration guide exists as a page but its body is the unwritten placeholder "To be written — API versioning policy / Differences between the current and previous version / Migration checklist / Deprecation timeline". deprecation: policy_url: null policy_published: false sunset_header: false deprecation_header: false note: >- No deprecation policy, no sunset timeline, and no RFC 8594 Sunset/Deprecation header support is documented. No `deprecated: true` operation appears anywhere in the OpenAPI. deprecated_operations: [] sla: published: false url: https://developers.kardinal.ai/reference/limits-and-quotas note: >- The limits-and-quotas reference lists SLA as "Not yet published — contact support@kardinal.ai to confirm for your account" for both sandbox and production. status_page: null status_page_note: >- No status page found. status.kardinal.ai does not resolve (NXDOMAIN) and no status/uptime page is linked from the website or the developer portal. changelog: url: https://developers.kardinal.ai/reference/changelog dated_entries: 0 note: >- The changelog page exists but contains only a template stub ("Reference — priority P2. Update this page with every notable API change.") and a single "Upcoming / To be defined" placeholder. No dated release has been published. detail: changelog/kardinal-changelog.yml surfaces: - name: Kardinal ARO API v2 state: live evidence: - {url: 'https://app.kardinal.ai/api/v2/public_key', http_status: 200, note: returns a real ES384 JWK} - {url: 'https://developers.kardinal.ai/openapi.yaml', http_status: 200, note: OpenAPI 3.0.3, 40 operations, 195 schemas} access: invitation only, per-customer environment, no self-serve sign-up - name: Kardinal self-serve "Route Optimization API" v1 state: announced-not-live announced_at: https://kardinal.ai/route-optimization-api/ claims: - POST /v1/optimize - 1,000 free stops per month, no credit card - from EUR 0.012 per stop on prepaid packs - Python and JS/TS SDKs - MCP server and agentic docs evidence_it_is_not_live: - {check: DNS, host: api.kardinal.ai, result: NXDOMAIN, note: the host its OpenAPI names as the production server} - {check: DNS, host: sandbox-api.kardinal.ai, result: NXDOMAIN, note: the host its OpenAPI names as the sandbox server} - {check: DNS, host: console.kardinal.ai, result: NXDOMAIN, note: the sign-up target linked from the landing page} - {url: 'https://kardinal.ai/pricing-route-optimization-api', http_status: 404, note: the "View pricing" link on the landing page} - {url: 'https://developers.kardinal.ai/openapi.json', http_status: 200, note: 'the v1 spec is a self-declared stub — info.description reads "Spécification à compléter avec les endpoints réels" (specification to be completed with the real endpoints), and it declares a single placeholder operation with untyped object schemas'} - {check: page-content, url: 'https://kardinal.ai/route-optimization-api/', note: 'the live landing page still carries unfilled template markers — "[MAX_STOPS]", "[N]+ constraints", and an HTML comment reading "A completer : LOGO_DARK_URL et les liens Services"'} disposition: >- NOT captured as an API in apis.yml and NOT saved to openapi/. The v1 spec is a scaffold the provider labels as incomplete, and none of the hosts it names resolve. Only the shipped ARO v2 API is profiled. Re-check on the next pass — if api.kardinal.ai comes up and the spec is filled in, this becomes a second API entry. documentation_maturity: note: >- The developer portal is actively under construction. Five reference/guide pages resolve to explicit "To be written" placeholders (concepts/data-security, concepts/sync-async-architecture, guides/migrating-api-versions, reference/changelog, and the `properties` section of reference/error-codes), and the OpenAPI itself carries "[TO_VALIDATE] Description pending review by a Kardinal engineer" on a number of schema descriptions. The pages that ARE written (error codes, limits and quotas, authentication, first API call, data model, agent modeling checklist) are unusually candid about what is not yet published.