local-deep-research · Issues· 898 open

report_generator: detailed-mode discards initial research synthesis and only passes 1000 chars to outline planner

hardening(security): release gate ignores note-level code-scanning alerts (zizmor class)

test(security): reword #4687's log_sanitizer example literals on landing (tripwire)

test: deflake sibling wall-clock perf-budget tests (pubmed / sqlcipher / alembic)

chore(ci): drop the actionlint $/ ignore once upstream ships support (rhysd/actionlint#711/#732)

feat(langgraph): emit a progress milestone when an oversized subtopic batch is rejected

perf(unified-search): skip redundant loading-spinner DOM rebuild on every keystroke

web: settings-page API-key save uses full force_refresh; research page's targeted refresh_provider is cheaper

web: model discovery gives no feedback for a wrong/unsupported API key

settings: static llm.provider options list is stale (missing all providers added after requesty)

providers: google/anthropic list_models overrides don't trim whitespace API keys

model dropdowns list non-chat ids (embeddings/rerank/multimodal) that fail at runtime

metrics: cost cards show $0.00 for QwenCloud (and DeepSeek/Atlas Cloud/Requesty) — no pricing entries

fix(security): bound the _URL_QUERY_STRING_RE path segment — quadratic backtracking on many-`://` scrub_error inputs

Add `normalize_url` to `embeddings.openai.base_url` for parity with LLM base URL