Skip to content

koinon 26.11.3

Initial knowledge provisioning

The installation workflow now prepares the native document and structured indexes through their required lifecycle states before recording readiness. For documents, the queued target enters the building state before the native PostgreSQL function creates its search index. Both domains must reach the ready state before installation can continue.

These transitions verify the tenant, workspace, target, model configuration and creating operator. Cancelled, retired, activated or changed targets remain protected. A cancellation observed during preparation prevents a ready checkpoint from being recorded. The PostgreSQL tests exercise the actual index functions, vector indexes, runtime roles and row-level security.

An interrupted installation retains its original plan and historical checkpoints. Inspect its receipt and explicitly resume that same plan using the initial knowledge provisioning guide. Updating the packages does not automatically restart the installation or resend administrator welcome messages.

Library availability

Library records distinguish recorded processing from observed availability. A successful extraction alone does not mark knowledge as searchable. The availability view checks the current native source, index and coverage; missing evidence or changed configuration remains unverified. Recorded attempts provide the processing history without changing historical receipts.

The same observations are available through the typed Python, TypeScript and Java SDKs. See the Library guide and knowledge service API.

File uploads and folder navigation

When queued storage is configured, the regular Library upload flow stages verified originals in durable storage and submits them for worker processing. An uncertain acknowledgement is checked through retained status rather than automatically sending the original again. Cancellation, staged originals, processing and search availability have distinct states. Deployments without queued storage retain the bounded synchronous path for smaller files.

The destination picker opens the relevant folder hierarchy without treating a top-level space as an upload folder. The Library root now requests only spaces from the catalog, so projects and their folders appear inside their actual parents while pagination remains server-side. Headless catalog queries retain their existing filters and behavior.

The chat and settings layouts share a bounded sidebar width and consistent content alignment. Settings can be collapsed, and the selected sidebar state is remembered. My activity opens the current user's activity from the navigation bar; organization-wide activity remains an administrator view. Browser routes now expose the existing communications and activity services through the session and CSRF protections.

The access and session-ended screens include an original isometric knowledge library, with responsive artwork for mobile and desktop. Custom organization branding retains its logo and a small Powered by koinon attribution. The sign-in action remains available when the decorative artwork cannot load. See identity and appearance.

Packages and upgrades

Use matching CalVer 26.11.3 distributions and their checksums from GitHub Releases. The runtime remains the official pyfly 26.10.32 release. Apply the normal migration and role checks before resuming any incomplete installation.