A rebrand month: a brand-new public website launched alongside a full visual
refresh of the web app, with big additions to albums, people, video, and
faces.
- New public website. A rebuilt marketing site launched at gumnut.ai, with new Home, Platform, Enterprise, About, and Contact pages.
- Rebranded web app. The web app adopted the Gumnut design system, including a new dark mode, a branded sign-in, and a new Gumnut-hosted sign-up page.
- More library management in the app. Create, rename, and delete albums; rename, merge, and set a featured face for people; single- and multi-photo downloads; move-to-trash; and faster parallel uploads with an ETA.
- Video in the viewer. Videos now play in the web asset viewer, with thumbnails across the timeline, grid, and MCP app, HDR tone-mapping, and a batch of paging fixes.
- New MCP tools.
view_assetreturns a photo’s pixels to the model for inspection, andsave_assetingests new photos and videos into a library. - Better face detection. A new SCRFD detector became the default, and you can now draw or delete face boxes by hand.
The biggest month yet, headlined by a brand-new Gumnut app and the first
release of video support.
- New Gumnut web app. A rebuilt, Vite-based single-page app launched at app.gumnut.ai, fully replacing and retiring the previous Next.js web client. It ships a fast, virtualized photo timeline with infinite scroll, an asset viewer with EXIF metadata, albums, people, search, uploads, API key management, and Immich and MCP setup pages. A matching mobile timeline landed in parallel.
- Basic video support. Videos now get thumbnails, duration, and streaming playback through the CDN (including HTTP range requests), with full end-to-end support through the Immich adapter.
- Trash and restore. Library-level trash/restore with an async purge window, plus Immich-compatible soft-delete so deletions sync correctly to Immich clients.
- Smarter face clustering. Switched to graph-based clustering (HNSW + Chinese Whispers), added crowd-face filtering, and shipped an operator face cleanup dashboard. Began migrating face embeddings to ArcFace.
- Closing Immich feature gaps. Map markers, “On this day” memories, and user edits to asset metadata (description, date/time, and GPS).
- Location data. Reverse geocoding now adds human-readable place names and location metadata to your assets.
- Face clustering quality. Density-aware gates, auto-merge, debouncing, and quality filtering significantly improved face grouping accuracy.
- CDN migration completed. Asset delivery through the CDN was finished end-to-end, including the Immich adapter.
- More Immich compatibility. People merge, faces, and reassignment endpoints; soft-delete (trash) backend; and compatibility with newer Immich releases.
- New app foundations. Scaffolding for the new web and mobile apps, including shared packages and the authentication flow.
- MCP and ChatGPT. MCP app refinements plus the groundwork for the ChatGPT
app submission (privacy policy, demo account, and the
merge_peopletool).
- AI image descriptions. Gumnut now generates descriptions for uploaded assets using provider-agnostic inference (OpenRouter and Google Gemini), with descriptions available through the API.
- CDN asset serving. Cloudflare-based CDN delivery rolled out across the API, web app, MCP app, and SDKs for faster image loading.
- MCP app stabilization. Theme support (dark/light), configurable CSP, iframe sizing, and hardened error handling.
- Search and people. Search and filter people by name, and a new endpoint to merge people.
- Performance. Faster deploys (~30 min → ~5 min), streamed uploads, and per-user rate limiting.
- Album covers. Albums now expose cover images and thumbnails.
- API improvements. Batch ID filtering on list endpoints, date-range filtering on assets, and full URLs returned directly in responses.
- Next-generation MCP. Completed client migration, deprecated the legacy MCP server, and added stateless HTTP mode for horizontal scaling.
- Faster faces. Face detection moved to a local task and images are downscaled first, speeding up processing.
- Reliability and security. First-class event modeling for sync, cursor pagination fixes, and cross-library access hardening.
- Next-generation Gumnut Photos MCP. Shipped the MCP foundation, the full Tier 1 toolset, complete MCP authentication, and tool enhancements (structured content, output schemas, titles).
- HEIC support. Fixed HEIC metadata and ML processing, and pre-generated full-size WebP so HEIC photos display in the browser.
- Immich adapter. WebSocket infrastructure with authentication, sync streaming, Immich-compatible error formatting, and an update to Immich v2.4.1.
- Immich adapter in production. Deployed the Immich compatibility layer,
adding sessions, checkpoints, and the
/sessionendpoints. - Sync foundations. A new events endpoint for client sync, plus SHA-1
checksum support and
device_id/device_asset_idfiltering for Immich clients. - Albums. Added start and end date fields to album responses.