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

# Changelog

> Product updates and improvements to Cognisafe.

<Update label="May 22, 2026" description="Weekly release">
  This week's release focuses on observability, search, and a refreshed marketing surface — plus a public on-host status page and roadmap so you can always see what's live and what's next.

  ## New features

  * **Real-time live feed.** A new live feed view streams requests through an animated pipeline with 3-second polling, so you can watch traffic and safety scores as they happen.
  * **Search and filter on Requests.** Filter request history by model, scorer, date range, and flag status to find incidents faster.
  * **On-host status page.** A public status page at `/status` reports live database and Redis health checks — no third-party dependency required.
  * **Product roadmap.** A new on-site [roadmap](https://cognisafe.uk/roadmap) shows what's shipped, in progress, and planned across the detect → prevent → automate → govern arc.
  * **Reference architectures.** Eight enterprise deployment patterns are now published — covering AKS, Helm, APIM, Azure OpenAI private endpoints, SIEM/Sentinel, and Grafana integrations. See [Self-hosting](/self-hosting/overview).
  * **Self-hosted LLM providers.** Added support for vLLM, NVIDIA NIM, Hugging Face TGI, and LM Studio alongside the existing [Ollama](/providers/ollama) integration — point the proxy at any OpenAI-compatible endpoint.
  * **In-product feedback widget.** A new feedback widget is available on the dashboard and marketing site so you can send notes directly to the team.

  ## Updates

  * **SEO and shareability.** Sitemap, `robots.txt`, OpenGraph cards, and JSON-LD metadata are now generated across the marketing site, so shared links render with proper previews.
  * **Sentry observability.** End-to-end Sentry error tracking is now wired into the API and web app, which means faster diagnosis and shorter incident response times.
  * **Proxy performance.** The proxy now uses a shared HTTP client for log delivery, reducing per-request overhead under load. No configuration changes required.
  * **Free-tier retention.** The retention purge job has been extended to clean up free-tier rows and cancelled account data automatically.
  * **Pricing tiers renamed.** Tier names are now Starter, Professional, and Business (previously Homelab, Pro, Team). Limits and features are unchanged.

  ## Bug fixes

  * **Feedback delivery.** Resolved multiple issues with the feedback widget, including misconfigured API URLs, missing error surfacing, and background task delivery.
  * **Encryption key length.** Corrected the development fallback encryption key to exactly 32 bytes (256-bit) to prevent startup errors in local environments.
  * **Retention timezone handling.** Fixed a timezone error in the retention purge that could skip rows near the cutoff boundary.
  * **Free-tier request limit display.** The homepage now correctly shows the 10,000 requests/month free-tier limit.
  * **Content Security Policy.** Corrected the `connect-src` domain so the feedback API loads cleanly in the browser.
  * **Authentication allowlist.** Public marketing and API routes are now correctly allowed through the auth middleware, fixing intermittent 401s on signed-out pages.
</Update>