> ## Documentation Index
> Fetch the complete documentation index at: https://docs.keplerinsights.us/llms.txt
> Use this file to discover all available pages before exploring further.

# Changelog

> Product updates to the Kepler Insights API.

## Week of 2026-05-12

### New

* **Public API launch.** The Kepler Insights scoring API is now generally available. Pick up an `ki_live_` key in the [developer console](https://console.keplerinsights.us) and start with the [Quickstart](/quickstart).
* **Sandbox environment.** Build and test against four canned test domains using `ki_test_` keys — no fetcher cost, no quota burn. See [Sandbox](/sandbox).
* **Five public tiers.** Free (14-day trial), Starter, Growth, Scale, and Enterprise, each with their own cached + cold call budgets and freshness windows. See [Tiers & limits](/tiers).

### Updates

* **Cold scoring is now always async on every paid tier.** When `POST /v1/score` needs to run a cold pipeline, the API returns `202 Accepted` with a `job_id` and `poll_url`. Poll `GET /v1/jobs/{job_id}` until `status` is `complete`. The Python and TypeScript SDKs handle the polling loop for you. See [Async scoring](/endpoints/async).
* **Sandbox responses are synchronous.** Sandbox `POST /v1/score` calls return the score inline as `200 OK` — sandbox does not queue jobs. See [Sandbox](/sandbox).
* **Freshness windows finalized per tier:** Starter 24h, Growth 6h, Scale 1h, Enterprise 15m. Cached responses within the window do not count against your cold budget.
* **Cold-spend circuit breakers.** Two automatic safety nets cap runaway cold usage: 200 cold calls/hour per key, and 2× your tier's monthly cold budget per account in any 24h window. Both return `429` with a `reason` field.

### Fixes

* `cache_status` no longer returns `"cold"`. The field is now `"cached"` or `"fresh"` — the cold-result case is delivered through the async job flow instead. See [`POST /v1/score`](/endpoints/score).
* The cohort endpoint now returns `no_history` (was `target_not_found`) when the target domain has never been scored. See [Cohort](/endpoints/cohort).
* Removed several error codes that the API never actually emits, including `400 async_requires_growth` and the spurious `502`/`504` rows. See [Errors](/errors) for the current, accurate list.
* The `wait` request field is accepted for backwards-compatibility but has no effect — cold scoring always uses the async job flow.