koinon 26.11.11

koinon 26.11.11

Organize your own Library

New installations start with an empty Library. Create the spaces, projects and folders that fit your team; koinon no longer creates predefined or protected knowledge folders. Existing folders and their contents remain in place. Authorized users can reorganize them through the ordinary catalog actions, while installation provenance remains immutable.

The Library fills the available page height, including empty and short lists. Its background context menu offers actions for the current location; item menus apply to the selected item. Available actions still follow the same server permissions. See the Library guide.

Add files and follow their progress

Select several files in one upload. koinon chooses each file's use from the installation's available capabilities: Word, PDF and PowerPoint files are prepared for content search, while supported spreadsheet and tabular formats such as Excel and CSV are prepared for row and column analysis. An optional per-file choice explains the effect as Search document content or Analyze rows and columns. There is no mandatory global technical mode.

The picker shows the destination, familiar file icons, readable KB/MB/GB sizes and rejected-file reasons. Format hints and effective limits come from the server's installed parsers. Transfer progress counts confirmed uploaded content; it does not invent a percentage for extraction or indexing.

You can close the upload panel or navigate within the app while uploads keep their original destination. Keep the browser tab open until transfer finishes. Submitted processing continues on the server; Uploads follows recorded status automatically and offers recovery only when applicable. Logout or an account/organization change clears local observation and prevents late responses from entering the next session. Failed and uncertain outcomes remain distinct, and an uncertain write is not automatically replayed. See formats and upload flow and durable intake.

The embedded MCP file review also uses readable, locale-aware sizes. Its single-file upload still requires explicit private approval; the documented Python, TypeScript and Java batch methods retain exact size/hash fields. See MCP upload and SDK workflows.

Validate Office files through the native readers

Bounded document intake rejects unsupported extensions and binary payloads disguised as text instead of falling back to a text parser. Capability descriptors use the admitted parser kinds and installed dependencies. Genuine DOCX, PPTX, XLSX and CSV fixtures exercise extraction and original-file preservation; spreadsheet/CSV readers also verify materialized rows and source coordinates. Legacy .ppt requires conversion with Save As .pptx in a compatible editor, not renaming the extension.

Original bytes, source versions and structured coordinates remain tied to their recorded receipts. Slide order is not a fabricated Word page number, and retaining images does not imply vision interpretation. The file ingestion reference describes the actual parser and resource limits. Local file intake does not provide SharePoint, OneDrive or Google Drive synchronization; that connector work remains separate.

Connect the desktop bridge through browser sign-in

The desktop MCPB source supports explicit browser OAuth with an operator-registered public client and an exact local callback. Sign-in starts only when the user requests it. Tokens stay in bridge memory; restart, expiry or revocation requires reconnecting. Additional scopes require consent, and the bridge never replays the original operation automatically.

Personal access tokens remain an explicit compatibility mode. Existing MCPB PAT configurations must select pat when upgrading. Server permissions remain authoritative in both modes. See desktop OAuth for registration, reconnect and revocation steps. Controlled protocol checks do not establish acceptance in an installed Claude client or a hosted connector.

Reduce repeated reads and qualify worker isolation

Knowledge and answer paths reuse bounded policy observations, batch catalog ancestry reads and validate audit memory at the final boundary. Conversation capacity is bounded separately from background-worker capacity. These changes reduce redundant work without relaxing authorization or claiming a deployment-specific response-time improvement.

PostgreSQL integration coverage exercises concurrent claims in separate worker processes, overlapping native document parsing, exclusion of a third claim at the configured limit, and cancellation fencing for one batch without replaying another. This is controlled worker qualification, not a cloud-scaling or production-capacity result.

Update the installation

Use 26.11.11 across runtime packages, canonical SDKs and interface assets. The runtime uses pyfly 26.10.32.

Apply native migration koinon_0043_ordinary_folders through the separate migration process before starting the updated application. It retires special folder protection while preserving existing content, identities and installation receipts. It does not delete folders or create a replacement default structure. An image-only update is insufficient.

Back up the database and rehearse the exact upgrade on a restored copy. Follow schema and role changes, then verify API and worker readiness, authorized folder operations, ingestion and an evidence-backed conversation. Preserve original files, source versions, conversation evidence and email delivery history; this release does not claim new email delivery or replay earlier messages.