v0.24.0

Released: October 7, 2026

Create and run Jupyter notebooks for Python projects.
CPU usage is now captured and exposed in the proxy, UI, and Prometheus metrics.
A unified process view displays all running processes in one place.
The loading page streams the starting worker’s log so you can follow startup progress.
On Kubernetes, the loading screen shows diagnostics for app startup failures.
A banner appears above every page when your license is approaching expiry.

Ricochet will automatically retry a failed app start up to five times.
The read API key scope is now enforced by all endpoints on the REST API.
Crashed apps now display their last log lines and link to the full runtime log.
The task overview page now informs of the next schedule run time.
Content type filters are now grouped by language.
Improved the UI interpreter display at /environment.
The execution environments page uses a single aligned grid layout.
Improved application worker scaling logic.
We improved the performance of log streaming.
Host interpreter discovery is skipped when a backend is already configured, improving startup time.
On Kubernetes, the readiness probe verifies database access before marking the service as ready.

Fixes a bug where cron schedules with 5 fields were not correctly registered after a server restart.
Tasks fail immediately with a clear error when Kubernetes refuses to create the pod.
Static content no longer serves source files when the output directory is set to the bundle root.
Scaling and schedule updates are rejected when applied to the wrong content type.
Ricochet falls back to community edition features immediately when a license expires or fails validation.
Entra authentication providers without a client secret are rejected at startup.
User change events are published when an admin changes a user’s role.
Existing renv cache entries are linked instead of replaced during restoration.
Stale workers no longer cancel retried renv restoration attempts.
Kubernetes restore jobs are reliably cleaned up when a restoration is cancelled or stale.
Ricochet waits for slow-starting containers to produce logs instead of timing out prematurely.
Request metrics are counted by content ID instead of slug.
Server-side rendering completes properly even after the client disconnects.
Tag input is preserved when appending to saved tags.
Mount discovery is serialized to prevent conflicts during container creation.