API status & changelog

What uptime and response time to expect, the limits that apply today, and how we roll out changes. The machine-readable version is at /api/status.json.

Current status: loading live status…

Latency & error SLOs

These are the service-level objectives we hold ourselves to. They are targets, not a contractual SLA: the API is in early access and provided as available. We alert on the objectives that have a live in-process signal (egress probe latency and consecutive extraction failures) and monitor the rest from request telemetry.

Objective	Target	Window	How it is evaluated
API availability Share of API requests that return a response other than a 5xx, over a 30-day rolling window.	`99.5% of requests succeed`	30-day rolling	non_5xx_ratio >= 99.5%
Response time — cached and metadata endpoints Cached transcripts, /api/health, /api/plans and other metadata reads. Liveness must never block on a third-party call.	`p95 ≤ 500 ms`	30-day rolling	p95 <= 500 ms
Response time — uncached extraction A first-time transcript extraction (YouTube caption fetch) for a single public video. Cached repeats are served from the response-time objective above.	`p95 ≤ 8 s`	30-day rolling	p95 <= 8 s
YouTube egress probe latency Latency of the background egress probe reported at /api/health and /api/health/egress. A rising probe is an early warning that extraction will degrade.	`p95 ≤ 2,000 ms`	rolling probe	probe latency <= 2,000 ms
Consecutive extraction failures Consecutive failed extractions before an operator alert fires. Any run of three is treated as a dependency incident, not a user error.	`0 (alert at 3)`	live	consecutive_failures < 3
Server error rate Share of API requests that return a 5xx. Expected extraction failures (blocked/unavailable videos) return a mapped 4xx and are not counted here.	`< 1% of requests are 5xx`	30-day rolling	5xx_ratio < 1%
Liveness response time GET /api/health and GET /api/health/ready must answer immediately; dependency state is reported, not awaited, on the liveness path.	`≤ 100 ms`	live	liveness responds <= 100 ms

Health endpoints: GET /api/health is a liveness check that always answers immediately and never blocks on a third-party call. GET /api/health/ready is a readiness check that returns 503 when the YouTube egress is unavailable, so orchestrators can take a broken instance out of rotation. GET /api/health/egress runs one bounded egress probe on demand.

Current limits

Free plan: 10,000 credits per account, 120 requests per minute per key.
Anonymous visitors: 600 credits per day per visitor, no key required.
Batch requests: up to 50 video IDs per POST /v1/transcripts call.
Credits are consumed per successful transcript; a batch authorizes its full cost before processing.
Higher limits are available for larger workloads — see /api/plans or contact us.

Rate-limit policy

Rate limits are per API key (or per visitor/IP for anonymous traffic) and are returned on every response:

Header	Meaning
`X-RateLimit-Limit`	The request ceiling for the current minute.
`X-RateLimit-Remaining`	Requests left in the current window.
`X-RateLimit-Reset`	Unix time when the window resets.
`Retry-After`	Seconds to wait, sent with a `429`.

When you exceed the limit the API returns 429 RATE_LIMITED with a Retry-After header. We recommend exponential backoff with jitter and honoring Retry-After. A burst above the limit is rejected, not queued.

Deprecation policy

Additive changes ship without notice. New endpoints, new optional parameters and new response fields are backwards compatible.
Breaking changes are announced here and in the changelog below at least 90 days before they take effect, with a migration path.
Deprecated endpoints keep working for at least 12 months after the announcement, and responses may carry a Deprecation or Sunset hint header.
Versioning. The API is served under /v1. A future incompatible version will be introduced alongside /v1 rather than replacing it in place, so existing integrations keep working.

Changelog

September 2026 — Published /api/status: latency/error SLOs, rate-limit and deprecation policy, and this changelog. Added a readiness probe at /api/health/ready and operator alerting for egress health regressions and runs of consecutive extraction failures. Live status is available at /api/status.json.
September 2026 — Batch transcript requests now authorize the whole batch cost up front, returning a clear INSUFFICIENT_CREDITS error instead of partially charging a request that cannot complete.
September 2026 — Translation fallback hardened: keyless auto-translated tracks that YouTube bot-gates were removed, and native-language translation returns real translations.
September 2026 — Signup hardened with a CAPTCHA challenge and multi-account abuse protection, plus a credit-exhausted upgrade nudge in the dashboard.
September 2026 — Developer API site redesign, an API blog in 10 languages, and dedicated legal and contact pages.
September 2026 — Consumer site redesign and the post-analysis video workspace.

Incidents

This page and /api/status.json are the source of truth for availability. If something looks wrong, include the X-Request-ID from the failing response and send us a message — that is usually enough to reproduce it.