Fix token display backgrounds: use themed card_bg instead of dark backtick code blocks for subnetwork & token analysis

Fix embedding tokens: soft blue instead of viridis black/green; make toggles, sidebar button, checkboxes more visible in light mode

Fix circuit graph visuals: remove node outlines, softer edges, theme-aware layer labels, reduce grid, black text for all labels/legend/ticks

Fix subnetwork tc UnboundLocalError, expand chat input container light-mode CSS

Fix circuit tc UnboundLocalError, chat input bar, heatmap/PCA/FV titles, highlight text, vline annotations black in light mode

Fix light mode: buttons, download btns, option_menu, heatmap/chart fonts, token colors in AI explanations

Theme-aware colors: replace all hardcoded dark colors with get_theme_colors() utility for proper light/dark mode

Proper light/dark mode: switch toggle, comprehensive CSS overrides for all inline dark colors, restore dark base

Add light/dark mode toggle with beautiful color schemes for both modes

Simplify chat header to 'Questions' / 'Fragen'

Fix chat truncation (max_tokens 800→2000) & clean up circuit trace key names

Fix FV context collector to use real data keys, silently skip missing influence tracer index

Fix influence tracer: use cached docs without pop, skip live index when missing; restore clear button without emoji

Enrich chat with background knowledge per page, remove clear-chat button

Update API key and migrate model from qwen2.5-vl-72b to qwen3-vl-30b-a3b-instruct

Fix follow-up chat: remove system role and top_p to match working API pattern, add error logging

Add interactive follow-up chat to all analysis pages with context awareness

Force dark mode theme for app

Fix indentation and add retry logic for failed circuit explanations

Disable generation controls for static mode

Update function vectors faithfulness logic and full cache

added examples cache

Add cached circuit trace results

Prevent caching of API errors in circuit trace analysis

Fix FV visualization errors and faithfulness caching

Add cached analysis results with LFS

Update full cache with explanations and faithfulness1

Update full cache with explanations and faithfulness

Fix context key error in attribution page

Fix Kaleido error with try-except and chromium package

Add vectors with LFS2

Deploy static demo (secrets removed + deps updated)

Deploy static demo