Add UI-configurable InfrastructureSettings with DB+env fallback for all features

Add 5 NotebookLM parity features: Mind Map, Infographic, Slide Deck, Deep Research, Video Generator

fix: use a real speech clip for STT connection tests (#838)

feat: complete audio matrix (Google/Vertex TTS, Google/ElevenLabs STT) (#835)

feat: expose new audio providers (Mistral STT/TTS, Deepgram TTS, xAI TTS) (#834)

feat: bump esperanto to 2.22.0 + Ollama num_ctx override (#833)

fix(embedding): drop degenerate tiny chunks before embedding (#764) (#768)

Make embedding batch size configurable (#742)

feat: use token-based sizing for embedding chunking (#749)

Merge pull request #753 from lfnovo/fix/graceful-credential-decryption-errors

fix: narrow exception handling and support migrate_to for broken credentials

fix: handle credential decryption errors gracefully (#740)

fix: map base_url to endpoint for Azure credentials (#741)

fix: prevent SurrealDB injection via order_by and unparameterized queries

feat: add DashScope (Qwen) and MiniMax provider support

fix: persist source asset, preserve custom titles, cascade-delete credential models

chore: bump version to 1.8.1

fix: narrow exception to (ImportError, OSError) and include error in log

fix: handle tiktoken network errors in offline environments (issue #264)

Upgrade default Azure API Version for testing and fetching models (#638)

feat(podcasts): model registry integration, credential passthrough & new features (#632)

fix: embedding batch sizing and 413 error classification (1.7.4)

docs: update documentation for error handling and podcast retry (#599)

fix: podcast failure recovery and retry (1.7.3) (#595)

fix: resolve merge conflicts and apply extract_text_content to all graphs

fix: restore graceful fallback in get_default_model and truncate error messages

feat: improve error clarity for LLM provider failures (#506)

fix: fail fast when source content extraction returns empty

fix: handle empty/whitespace source content without retry loop (#576)

fix: update esperanto dep and increase transformation max_tokens (#568)

fix: turn the embedding field into optional (#557)

feat: credential-based API key management (#477) (#540)

fix: handle structured content format in LLM response parsing

feat: add environment variables for chunk size configuration (#520)

fix: use sync get_state() for SqliteSaver compatibility (#519)

Fix re-embedding issues and improve retry strategy (#515)

fix: async insight creation to prevent transaction conflicts (#512)

feat: use standard HTTP_PROXY/HTTPS_PROXY environment variables (#499)

feat: message counting for chat sessions (#430)

fix: improve error handling in repo_create and repo_insert (#474)

feat: add cascade deletion for notebooks with delete preview (#471)

fix: improve error logging for chat model configuration issues (#458)

feat: content-type aware chunking and unified embedding (#444)

perf: improve source listing speed by 20-30x (#436)

Feat/localization tests docker (#371)

fix: delete uploaded files when sources are removed

fix: reduce retry log noise during concurrent chunk processing

feat: improve dev commands, update all langchain dependencies to their latest major versions

docs: generate comprehensive CLAUDE.md reference documentation across codebase

refactor: reorganize folder structure for better maintainability