Biorouter v1.93.0

Biorouter v1.93.0

This minor release brings a simpler desktop interface and website, improves local inference and bug reporting, and checks built app pages before reporting a successful build, launch or export.

Desktop

• The composer, chat transcript, tool results, previews, Settings and component pages now share compact controls, consistent typography and keyboard-accessible help. Biorouter retains its coral accent and light/dark theme families.
• Chat history can be grouped by date or folder and sorted by activity, creation time or name. Rename is available from the row menu and keyboard. The chat summary docks beside the conversation when space permits.
• Window resizing and preview growth preserve page geometry. Settings section links and scrolling keep the first rows below the top bar. Reduced motion is respected.
• Crew moves into Settings > Chat > Capabilities and keeps your existing on/off choice. It no longer appears as an extension or in the Tools popover.
• Extension failure notices replace duplicate failure toasts while the Extensions page is open. Tool switches have accessible names, and development launches display Biorouter's own version.

Local inference and scheduled work

• Ollama uses its native chat API with an explicit context window and truncation disabled. Context overflow is reported instead of silently shortening the prompt. Private sessions refuse cloud alternates and remote aliases before inference.
• Managed Llama Server verifies runtime/model downloads, authenticates startup and defaults to one request at a time. Automatic context selection can retry an out-of-memory startup at a smaller window; persistent failures show guidance.
• Unambiguous tool calls written as text can be recovered when their names and arguments match the available tools. Recognized tool-call-only replies that cannot run get a notice that nothing ran. Recovery does not execute repaired arguments or ordinary prose examples.
• Ollama idle timeouts name the setting to adjust, including a response that sends headers and then stalls. The failed prompt is sent once.
• Daily Meditation upgrades older shipped workflows and resolves the former soul-writer skill name. Runs that fail before a message leave no empty chat; an aborted run preserves the meditation cursor.
• GPT-5.6 Sol is now the default for Versa API Azure.

Built apps

• Browser-free page checks run by default for build, launch, serving and export. They check the entry, declared local scripts/styles and JavaScript structure. Broken staged output does not replace the previous working bundle. A lint override cannot waive a hard page defect.
• App agents may use extensions installed and enabled on the running machine, with that machine's saved configuration and existing permission/privacy rules.
• Exported shell and Node launchers reuse the verified service for the same profile on macOS/Linux when available, and leave it running when the launcher exits. Explicit daemon settings retain precedence. Windows retains explicit configuration and standalone startup; shell automatic discovery requires Python 3.
• Built-app opening, fallback bundling and repeated HTML injection are corrected.

Bug reporting

• /bug is available in the desktop, TUI and classic CLI. Reports include an evidence-based suspected cause and check intended behavior against the documentation and source.
• Private chats require human approval and open a prefilled GitHub page for you to review and submit. They do not automatically upload the report or diagnostics bundle.

Validation limits

A browser-free page pass does not establish rendering, interaction or runtime behavior. The reporter's small Ollama models still failed to complete the clinical tasks in bounded replays. The existing CLI error-turn exit status and fixed prompt size for 32k-window models remain unchanged.

Crew user manual