PeakD Image Proxy Experiment: Where We Are and How We Got Here

Let's take a break from [AI tooling](https://peakd.com/hive-139531/@asgarth/a-hive-ai-agent-skill-and-a-modern-cli-built-for-the-ai-age) [stuffs](https://peakd.com/hive/@asgarth/ai-agent-tooling-update-new-cli-features-and-more-capable-hive-skill) and let's focus on some potential improvements for PeakD and Hive: an alternative image proxy project we started for PeakD. The short version is simple: - This started as an experiment because of recent issues with the default Hive image server PeakD depends on. - This is still exploration and brainstorming, not a drop-in replacement we can swap into production overnight. - The project is already useful, but we are still validating behavior under real load and edge-case traffic.  ## Why this exists PeakD is image-heavy, and instability on the default Hive image server directly impacts user experience. Instead of only patching around incidents, we started testing a PeakD-focused proxy pipeline as an alternative option to use in the future. Important: this is still an experiment. It is not guaranteed we will be able to switch PeakD to this new system. ## Architecture in brief The evolution so far: - Started from a lite port of `hive/imagehoster` (https://gitlab.syncad.com/hive/imagehoster) - Removed features not needed for PeakD (upload, filesystem storage) - Reowrked the code to use multiple concurrent processes - Evolved to `dispatcher + queue + workers` for explicit backpressure control Current stack: - Bun + TypeScript - Hono as web framework - cache-first request flow - bounded queue + worker pool + overflow fallback - status endpoint with metrics and worker health ## Current status Load and performance testing are ongoing. Testing that the images are loaded is the easy part. The difficult part is tuning the queue/concurrency, validating cache dynamics, and testing failure modes under pressure. ## Next steps After more tests, we can start controlled experiments on PeakD. Even with full API compatibility, PeakD itself still needs a small integration rework to route traffic safely: - A/B testing controls - gradual traffic shifting - safeguards to avoid moving the whole community at once and overloading the new proxy Project repository (public, work in progress): https://gitlab.com/peakd/peak-image-proxy If this direction proves stable enough, an official post from `@peak.open` may follow.