Local Deep Research

Highlights

1.8.0 is a large release weighted toward security hardening — 36 security changes, 65 bug fixes, 17 new features, 4 breaking changes, and 7 removals.

Most of the security work falls into two areas:

• Egress policy. LDR controls where research traffic is allowed to go — via PRIVATE_ONLY / PUBLIC_ONLY / ADAPTIVE scopes, explicit per-call policy checks, and a process-wide PEP-578 socket.connect audit-hook backstop for paths the explicit checks don't cover. Snapshot-less programmatic callers now fail closed instead of constructing a cloud client, cloud-metadata/SSRF endpoints are blocked more thoroughly (including alternate IP encodings and metadata.* hostnames), and operator-supplied endpoint URLs are SSRF-validated before the SDK runs. Note: the scope model is being reworked into a two-axis classification (see #4882 / ADR-0006), so expect refinements in a later release.
• Credential leaks. The unrecoverable SQLCipher master password is now scrubbed from tracebacks and kept out of every durable or shipped log sink; the settings endpoints redact API keys; and the settings form no longer pre-fills stored secrets. Separately, a FAISS pickle-deserialization RCE is closed and WebSocket connections default to same-origin.

Breaking changes to check before upgrading:

• The mcp / agentic (ReAct) and auto / parallel search strategies are removed; existing selections migrate automatically to langgraph-agent.
• Programmatic get_llm(provider=…) / get_embeddings(provider=…) fail closed without a settings snapshot (raise PolicyDeniedError).
• The default egress scope is now adaptive; set Egress Scope to Both to keep the previous "any engine, cloud inference" behavior.

Full list below — each item links to its PR.

What's Changed

🔒 Security Updates

• fix(settings): scope settings_changed WebSocket event to the owning user (cross-user leak) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4437
• fix(search): stop broadcasting search socket events to all connected users by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4446
• feat(security)!: egress policy module + search/LLM/embeddings PEPs by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4300
• fix(security): detect URL-encoded and unicode-traversal attacks in PathValidator by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4291
• fix(security): green main (TestQuickSummary) + clear egress block messages + quick_summary settings opt-in by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4456
• ci(release): add three cheap sanity checks (twine, tag-on-main, SBOM) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4225
• fix(security): arm egress audit net for direct MCP searches; fix library engine settings reads by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4517
• fix(security): bump base image to python:3.14.6-slim, refresh .grype.yaml by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4523
• ci(puppeteer): install test deps with npm ci instead of npm install by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4524
• fix(policy): align egress scope fallbacks with registered default by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4465
• fix(policy): use static engine class flags for egress scope decisions by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4471
• fix(security): redact SQLCipher password in encrypted_db, scheduler, and queue failure logging by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4530
• fix(security): add runtime egress scope verification in BaseSearchEngine by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4472
• fix(history): pass username explicitly to get_research_strategy (follow-up to #3661) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4526
• feat!: remove the auto and parallel meta search engines by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4534
• docs(egress): add runtime backstop + MCP audit net to enforcement map by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4543
• test(history): lock username as required keyword-only on strategy functions (follow-up to #4526) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4541
• test(security): AST invariant — no traceback logs with password in scope by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4540
• fix(security): redact + break chain on DatabaseInitializationError re-raise (#4182) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4542
• fix(security): redact password leak in DatabaseInitializationError consumers (#4182) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4550
• fix(security): redact password leak in ThreadSafeMetricsWriter.get_session (#4182) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4553
• fix(security): parse LDR_TEST_MODE as a boolean for the KDF floor by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4564
• fix(security): close password frame-locals leak class at the sink + targeted sweep (#4182) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4555
• test(db): lock in LDR_TEST_MODE boolean parsing for the KDF floor by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4565
• fix(security): force diagnose=False on the file sink too (#4182) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4568
• fix(security): extract sanitize_error_message to standalone function + strengthen patterns by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4466
• feat(benchmark): record LDR version + redacted settings snapshot at benchmark start by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/3844
• feat(security): validate LLM provider base_url against SSRF rules by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/3929
• refactor(rag): categorize /test-embedding errors instead of guessing (#4208 follow-up) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4220
• chore(security): SSRF housekeeping cleanups by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/3931
• fix(llm): address audit findings from PR #3670 by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/3767
• fix(search): strip URLs before the _is_valid_search_result gate (searxng + mojeek) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4573
• fix(settings): close password-input HTML leak + protect against [REDACTED] write-back by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/3954
• fix(chat): guard chat research against missing DB password (#4457) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4571
• refactor(security): centralize error scrubbing in BaseSearchEngine._scrub_error helper by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4572
• fix(settings): redact remaining GET endpoints + extend empty-password guard to /save_settings by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/3988
• fix(benchmark): gate evaluator endpoint_url behind the settings opt-in by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4579
• fix(benchmark): quote + escape free-form string fields in YAML export by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4583
• security: add rel=noopener noreferrer to external target=_blank links by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4142
• fix(security): resolve open code-scanning alerts (pypdf DoS, exception exposure, migration SQL, Actions injection, devskim FP) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4601
• fix(deps): patch js-yaml + @babel/core Dependabot alerts in test tooling by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4614
• security: enforce rel=noopener noreferrer via pre-commit hook by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4144
• fix(security): bound the RAG auto-index queue to prevent OOM under burst by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/3939
• chore: remove dead SearchCache utility, deprecate orphaned table by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4621
• fix(library): clamp documents API limit/offset to prevent unbounded load (#4560) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4596
• security: flag dynamic target=_blank links in check-target-blank-rel hook by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4626
• fix(history): paginate /api/history to bound memory (#4560) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4597
• fix(library): scrub credentials from download-service errors before they reach the client by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4625
• feat(llm): Anthropic-compatible (Messages API) custom endpoint provider (#4559) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4581
• fix(security): scrub credentials from search/fetch/agent tool errors (follow-up to #4625) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4633
• fix(deps): patch cryptography (HIGH) + aiohttp CVEs (Trivy #7877-#7885) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4634
• fix(deps): patch crawl4ai/langchain/starlette/python-multipart CVEs (Trivy/Grype wave) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4642
• refactor(settings): drop dead hand-rolled model-listing, single discovery path (H7) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4646
• fix(security): widen credential-scrub patterns (OAuth/Azure/Google keys, auth headers) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4640
• fix(security): close FAISS pickle-deserialization RCE with a restricted unpickler by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4632
• security: harden backup SQL injection guard and validate glob results by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/3098
• fix(ci): allowlist unfixable js-yaml DoS in the npm-audit release gate by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4651
• security: apply glob hardening to backup-status API route (follow-up to #3098) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4663
• test(db): fix encrypted-DB test that silently skipped in CI (wrong package name) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4674
• fix(deps): patch crawl4ai RCE (CRITICAL x2) + langsmith/pydantic-settings/pypdf CVEs by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4681
• security: harden warning_checks backup glob + make is_safe_glob_result shared by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4682
• fix(security): scrub GitHub/AWS/Slack/Google-OAuth tokens + JWTs from error text by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4664
• fix(security): escape innerHTML injections across analytics pages by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/3095
• chore(gitleaks): move password_utils + search_engines_config to path allowlist by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4733
• chore(deps): bump actions/checkout from 6.0.3 to 7.0.0 by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4738
• fix(egress): single source of truth for a run's primary engine (fixes mid-run scope_mismatch crash) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4745
• fix(langgraph): propagate search context (DB password) into subagent pool by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4751
• fix(egress): route RAG-indexing + SBERT embeddings PEPs through resolve_run_primary_engine by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4753
• fix(egress): close inference-path fail-open (LLM/embeddings PEPs use the single primary source) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4752
• fix(fetch): return a recoverable tool message on a per-URL egress denial by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4761
• fix(security): ignore CVE-2026-48931 in patchright-bundled Node (alert #7994) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4811
• fix(ci): fail loudly when Grype produces no SARIF instead of uploading an empty one by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4796
• ci(security): ignore unfixable CVEs in Grype by fix state by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4797
• fix(security): repair non-functional Bearer XSS/redirect suppressions in JS (9 false positives) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4799
• ci(release-gate): wait for grype-scan before checking code-scanning alerts by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4817
• fix(security): green the WebSocket-CORS unit suite + fix stale '*' default docs (#3091 follow-up) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4807
• fix(security): rework #3090 symlink + LDR_DATA_DIR hunks (500 on non-root, apostrophe crash) + add tests by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4808
• fix(search): restore TinyFish ruff-format gate + redact query from request-error logs (#4057 follow-up) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4806
• ci: fail loudly when Bearer/Semgrep produce no SARIF instead of uploading an empty one by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4818
• fix(security): make Bearer SQLi suppressions in migrations 0013/0008 actually work by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4792
• fix(security): don't log the user's path when blocking a restricted directory by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4820
• feat(security): warn at startup on weak SQLCipher KDF with existing user DBs by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4816
• build(security): pre-commit guard for malformed bearer:disable directives + cleanup by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4821
• fix(security): don't log the user's submitted path on a RAG path-validation failure by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4825
• fix(security): log folder/file basenames (not full paths) in RAG local indexing by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4831
• fix(notifications): surface validator reason + admin hint in test-URL error by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4002
• fix(security): don't leak URL-parse exception in notification test error (CodeQL #4775) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4835

💥 Breaking Changes

• remove(benchmarks): drop cross-run result reuse by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4498
• refactor(llm-providers): collapse dual-path LLM construction; fix Ollama enable_thinking by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/3984
• chore: remove the dead /research/api/config endpoint by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4551
• refactor(journal_quality)!: remove the retired DOAJ Seal quality signal by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4531
• refactor(routes): extract _research_not_found helper (DRY) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4713

✨ New Features

• Remove experimental search strategies not in the frontend by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4420
• feat(library): per-collection 'available to the research agent' toggle by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4453
• feat(benchmarks): add sampling seed for reproducible benchmark question selection by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4516
• ci(release-gate): CPU baseline smoke test under qemu-emulated CPUs by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4506
• fix(api): load user context for REST research endpoints (and remove /quick_summary_test) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/3661
• feat(library): warn in research log when collection search results are incomplete by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4520
• feat(warnings): add action link to model-mismatch hardware warning by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4529
• ci(ai-review): auto-run on every push to non-draft PRs by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4554
• ci(forensics): request arrival/duration logging for #4431 navigation-stall hunts by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4536
• feat(dev): add run_test_instance.sh launcher for a disposable LDR dev server by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4382
• feat(ci): run AI code review across a configurable list of models by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4667
• feat(search): cap local RAG results (default 20) + opt-in LLM relevance filter by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4765
• feat(search): group the search-engine selector into ordered bands with headers by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4814
• feat(metrics): a11y empty states + link feedback to source research by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4793
• feat(library): opt-in scheduled sweep to index unindexed library documents (#3939 follow-up) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4627

🐛 Bug Fixes

• test(conftest): stop QueueProcessorV2 thread between tests by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4441
• fix(ui): bind listeners before async init awaits in pdf-upload + help by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4421
• chore(news): drop redundant inline smooth-scroll override by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4447
• test(downloaders): fix GC-timing flake in context-manager close test by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4444
• fix(tests): instant scroll in mobile-nav overlap test (defeat Bootstrap smooth-scroll) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4442
• fix(library): make FAISS quarantine WARNING actionable by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4221
• fix(library): make ODT and office-doc uploads actually work (#4414) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4415
• fix(progress): rebalance progress bar for detailed reports by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/3806
• fix(search): validate and deduplicate follow-up relevance filter indices by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4337
• fix(news): honor flat source_type fields in SQLCardStorage.create() by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4245
• fix(news): subscription folder dropdown never loads and folder selection is dropped by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4495
• fix(embeddings): import text splitters from concrete submodules, not package root by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4493
• fix(deps): pin faiss-cpu to 1.13.x to avoid SIGILL on non-AVX2 CPUs by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4491
• fix(search): honor search.max_results in Library/Collection RAG engines by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4496
• fix(benchmarks): make xbench-DeepSearch respect the configured number of examples by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4494
• fix(loaders): clear error for encrypted .xls files by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4467
• fix(research): refuse LLM synthesis when no sources are available (prevents fabricated citations) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4503
• fix: extract WARNING_AFFECTING_KEYS constant + add missing egress keys (DRY PR 6/23) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4463
• fix(tests): make QueueProcessorV2.stop() interrupt the loop wait — removes ~10s teardown per app test by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4533
• fix(api): 400 for unknown /analyze_documents params + document allow_default_settings by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4527
• fix(scaleserp): guard against null link dropping entire result set by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4528
• fix(tests): disable Chrome leak-detection dialog that ate CDP input after login (#4430) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4501
• fix(ui): show actual search engine name instead of "web_search" by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4470
• test(api): real-path regression tests for /generate_report user-context (#4396) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4552
• fix(ci): honour fast SQLCipher KDF in test servers via LDR_TEST_MODE by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4558
• fix(metrics): record LLM calls when provider reports no token usage by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4473
• fix(settings): collapse sections by default on mobile by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4032
• fix(library): don't quarantine healthy FAISS index when embedding provider is down by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4512
• fix(ui-tests): real assertions for collection delete + subscription toggle/delete via API seeding by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4174
• fix(ui): render friendly engine name in agent-thinking tool_call panel (follow-up to #4470) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4570
• fix(rag): dedup collection indexing across SSE route and background worker by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4502
• fix(ui-tests): drop dead-field subscription tests, retarget at #subscription-strategy by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4151
• fix(ui-tests): real assertions for document upload + delete + bulk-select by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4187
• fix(news): make subscription status the single source of truth for active/due by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4492
• test(settings): regression-test bulk warning recalc for egress keys (#4463) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4582
• fix(db): don't load report_content/chunk_text into memory in list queries (#4560) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4574
• fix(settings): remove duplicate LLAMACPP entry from model-provider dropdown by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4594
• fix(metrics): drive rate-limiting analytics from RateLimitEstimate (dead panel) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4576
• fix(models): use_alter on ResearchResource.document_id to break circular FK by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/3776
• fix(library): batch convert_all_research to bound memory on large history (#4560) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4585
• refactor: remove redundant per-site -stripping (rely on central wrapper) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4343
• test(ui-tests): real assertions for 10 settings-interaction tests (dead-field audit PR 1) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4589
• test(themes): generate theme CSS from registry instead of the gitignored file by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4599
• fix(settings): restore shared_library safety warning by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/3766
• fix(llm): guard central think-strip against non-string content (follow-up to #4343) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4611
• refactor(llm): make auto-discovery the single source of truth for providers by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4591
• test(ui-tests): real assertions for 4 settings-pages Input tests (dead-field audit PR 2) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4613
• fix: add thread safety to SearchCache memory cache by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/3138
• fix(news): correct subscription run-now schedule + repair broken folder PUT route (#4492 follow-up) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4587
• test(ui-tests): real assertions for library collection-card + doc-search (dead-field audit PR 3) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4617
• fix(settings): source the provider dropdown from auto-discovery (not a hardcoded list) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4622
• fix(rag): make H3 embedding_dimension probe real + extend dedup to index_all (#4502 follow-up) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4593
• fix(news): repair subscription-history and organized-subscriptions 500s by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4588
• fix(memory): cap persisted log message size + share /history/logs cap constants by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4240
• Fix AttributeError on list-type LLM content (recovered from #4628) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4644
• fix(logs): honor ?limit on /api/research//logs by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4645
• fix(ui-tests): make escapeKeyFunction test real app behavior (was failing with navigation error) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4655
• fix(logs): tie-break equal-timestamp log ordering by ResearchLog.id by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4661
• fix(errors): preserve tailored error messages in error report generator by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4087
• test(ui-tests): real assertion for semanticColors (dead-field audit) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4658
• fix(errors): stop over-broad error patterns from giving wrong advice by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4666
• Drop dead Document.notes legacy column (recovered from #4619) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4643
• fix(ui-tests): make focusIndicatorVisible reliable in the accessibility shard by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4683
• fix(ui-tests): chat report-refactor test no longer closes the page without an LLM (shard fix) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4680
• test(ui-tests): real assertions for context overflow (dead-field audit) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4671
• fix(db): serialize per-user DB cold-open to prevent concurrent-migration race by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4665
• fix(research): detailed reports bypass report.enable_file_backup (H2) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4684
• fix(export,news): malformed RIS export + dropped subscription-history metadata by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4435
• fix(library): bound research-list dropdown query (#4560) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4719
• refactor(engines): unify API key placeholder list (DRY) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4715
• fix(library): stream get_unique_domains scan to bound memory (#4560) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4720
• fix(ci): stop unrelated labels from cancelling in-flight AI reviews by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4697
• fix(library): defer text_content in collection document loads (#4560) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4598
• test(ui-tests): real assertions for results star-rating (dead-field audit) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4670
• test(ui-tests): real assertions for news feed (dead-field audit) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4672
• test(ui-tests): real assertions for followup research (dead-field audit) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4673
• test(ui-tests): real assertions for error handling (dead-field audit) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4678
• test(ui-tests): real assertions for mobile interactions (dead-field audit) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4679
• fix(library): project ResearchResource whole-table scans to bound memory (#4560) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4724
• test(chat): de-flake per-user cap test (cleanup_middleware 1% sample) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4688
• test(settings): remove stale duplicate cleanup rate-limit tests (#4735 follow-up) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4741
• fix(research): carry username in run snapshot so the agent can use document collections by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4746
• fix(ci): whitelist vulture false positives blocking v1.8.0 release by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4755
• fix(ci): seed link-analytics shard data + bump experimental badge to 12px floor by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4763
• fix(tests): update tests stale after #3091 (CORS) and #4057 (TinyFish) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4815
• fix(ci): standardize server-readiness probes (curl -fsS /api/v1/health) + make responsive-ui cleanup real by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4794
• ci(pre-commit): retry only hook install, not the lint run by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4826
• fix(metrics): announce the real star-reviews load error; drop dead re-announce by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4828
• fix(journal-quality): follow OpenAlex 2026-06 standard-format snapshot layout by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4830
• fix(startup): defer heavy ML imports off the app-boot path (fixes UI test gates) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4829
• fix(web): defer WeasyPrint import so it no longer blocks server cold start (#4431) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4734

⚡ Performance Improvements

• ci(release-gate): cache the cpu-smoke pdm venv, pin two rationale comments by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4525
• test(ui-tests): replace delay(1000) with polling in context-overflow tests by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4708
• test(startup): add create_app-level import-weight regression guard by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4833

🗄️ Database Changes

• chore: deprecate (but keep) dormant UploadBatch model by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4178
• chore(db): drop orphaned cache tables, remove dead Cache/SearchCache models by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4641

📚 Documentation

• docs: add test-review (PUNCHLIST) completion report by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4293
• docs(test): clarify test_handles_errors_gracefully is intentional behavior by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4292
• test: annotate 28 PUNCHLIST KEEP-marked tests with audit-reviewed markers by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4295
• fix(settings): loguru brace formatting + docstring clarity in _filter_editable_settings by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4485
• docs(readme): document the x86-64 AVX minimum CPU requirement by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4513
• docs(progress): cross-reference _REPORT_PHASES emitters by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4481
• docs(resource-cleanup): record that asyncio.run loops don't reproduce #3816 by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4143
• docs: mark get_available_providers cache note as removed in #4590 by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4595
• docs: fix stale ai-reviewer.sh location in AI reviewer setup guide by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4718
• docs(library): document FAILED/COMPLETED revival in queue_research_downloads by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4696
• docs(changelog): note the orphaned cache-table drop (#4641) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4727
• docs: regenerate CONFIGURATION.md (add missing TinyFish + RAG rows) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4819
• docs(adr): ADR-0005 — reject in-app response compression by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4836

🔧 CI/CD & Maintenance

• chore: clear changelog fragments for 1.7.0 by @github-actions[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4440
• test: annotate 276 PUNCHLIST entries as 'reviewed — issue resolved by prior PR' by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4296
• chore: delete orphan web/static/css/auth.css by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4148
• chore(deps): bump actions/checkout from 6.0.2 to 6.0.3 by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4477
• chore(deps): bump github/codeql-action from 4.36.0 to 4.36.2 by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4478
• ci: cap pytest-tests job at 75 minutes by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4514
• 🤖 Update dependencies by @github-actions[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4509
• ci: raise pytest-tests timeout to 120 minutes by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4522
• refactor: delete dead files (DRY PR 3/23) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4460
• chore(ui-tests): drop dead-field history CRUD tests (unseedable in CI) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4580
• 🤖 Update dependencies by @github-actions[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4647
• chore(deps): bump zizmorcore/zizmor-action from 0.5.6 to 0.5.7 by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4739
• chore: bump minor version to 1.8.0 by @github-actions[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4743
• ci(pre-commit): lint that release-gate SARIF scanners are in both needs lists by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4822
• 🤖 Update dependencies by @github-actions[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4759
• test(ci): surface login HTTP status in register_ci_user diagnostics by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4791
• refactor(chat): remove dead/inert context code + orphaned max_context_messages setting by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4350

⬆️ Dependencies

• chore(deps): bump marked from 18.0.4 to 18.0.5 by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4475
• chore(deps-dev): bump happy-dom from 20.10.1 to 20.10.2 by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4476
• chore(deps): bump dompurify from 3.4.8 to 3.4.9 by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4504
• chore(deps-dev): bump eslint from 10.4.1 to 10.5.0 by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4606
• chore(deps-dev): bump eslint from 10.4.1 to 10.5.0 in /tests/puppeteer by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4603
• chore(deps): bump dompurify from 3.4.9 to 3.4.10 by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4604
• chore(deps-dev): bump happy-dom from 20.10.2 to 20.10.3 by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4605
• chore(deps): bump js-yaml from 4.1.1 to 4.2.0 in /tests/puppeteer in the npm_and_yarn group across 1 directory by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4608
• chore(deps-dev): bump js-yaml from 4.1.1 to 4.2.0 in /tests/api_tests_with_login in the npm_and_yarn group across 1 directory by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4609
• chore(deps): bump js-yaml from 4.1.1 to 4.2.0 in /tests in the npm_and_yarn group across 1 directory by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4610
• chore(deps): bump @playwright/test from 1.60.0 to 1.61.0 in /tests/accessibility_tests by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4629
• chore(deps-dev): bump @playwright/test from 1.60.0 to 1.61.0 in /tests/ui_tests/playwright by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4630
• chore(deps): bump puppeteer from 25.1.0 to 25.2.0 in /tests/puppeteer by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4750
• chore(deps): bump puppeteer from 25.1.0 to 25.2.0 in /tests/ui_tests by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4748
• chore(deps-dev): bump puppeteer from 25.1.0 to 25.2.0 in /tests/api_tests_with_login by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4747
• chore(deps): bump puppeteer from 25.1.0 to 25.2.0 in /tests by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4749
• chore(deps-dev): bump @playwright/test from 1.61.0 to 1.61.1 in /tests/ui_tests/playwright by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4757
• chore(deps): bump @axe-core/playwright from 4.11.3 to 4.12.1 in /tests/accessibility_tests by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4756
• chore(deps): bump @playwright/test from 1.61.0 to 1.61.1 in /tests/accessibility_tests by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4758
• chore(deps): bump puppeteer from 25.2.0 to 25.2.1 in /tests by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4772
• chore(deps): bump puppeteer from 25.2.0 to 25.2.1 in /tests/puppeteer by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4771
• chore(deps-dev): bump puppeteer from 25.2.0 to 25.2.1 in /tests/api_tests_with_login by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4769
• chore(deps): bump puppeteer from 25.2.0 to 25.2.1 in /tests/ui_tests by @dependabot[bot] in https://github.com/LearningCircuit/local-deep-research/pull/4770

🧹 Code Quality & Refactoring

• refactor(settings): extract _filter_editable_settings helper by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4469
• chore(benchmarks): remove dead sampling paths from xbench dataset loaders by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4518
• chore(benchmarks): collapse identical dataset branches in _create_task_queue by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4519
• refactor: remove dead-code overrides (DRY PR 1/23) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4458
• refactor: use shared LANGUAGE_CODE_MAP in SerpAPI, Brave, Google PSE (DRY PR 4/23) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4461
• refactor: extract _extract_display_link helper (DRY PR 5/23) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4462
• chore: add Session type hints to db_session parameters across remaining files by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4487
• refactor: remove dead SEARCH_SNIPPETS_ONLY checks (DRY PR 2/23) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4459
• refactor(benchmark): extract YAML export helpers into a unit-tested module by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4584
• refactor(llm): remove dead provider-availability surface from llm_config by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4590
• refactor(auth): centralize research DB-password guard in password_utils by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4577
• chore(ui-tests): remove dead-field orphan formSubmitWithQuery (dead-field audit) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4638
• refactor(api): single-source the Ollama /api/tags probe (H1) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4650
• refactor(auth): extract session create/cleanup helpers (DRY) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4712
• refactor(config): extract _ensure_dir helper in paths.py (DRY) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4714
• fix: correct test quality bugs and env var pollution by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/2976
• refactor(strategies): add base error-response and citation helpers (DRY) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4711
• refactor(settings): add unwrap_setting utility (DRY) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4716
• refactor(search): extract run_parallel_searches helper (DRY) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4717
• refactor(engines): extract _create_journal_filter helper (DRY) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4710
• cleanup: remove dead code and consolidate remove_think_tags duplication by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/2894

🧪 Tests

• test(logpanel): 20s timeout for the MAX_LOG_ENTRIES prune test by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4443
• test(ui): add CI coverage for journal quality dashboard and download manager by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4449
• test(journal_quality): add unit tests for db.py accessor methods by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4464
• test(api): endpoint-completeness meta-test + autospec real-path factories (follow-up to #3661) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4521
• test(chat): de-flake per-user cap test (remove 30s wall-clock dependency) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4549
• test(providers): guard that local providers stay auto-discovered by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4618
• test(ui-tests): real assertion for news subscription card (dead-field audit PR 4) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4624
• test(ui-tests): real assertions for mobile dropdown-open + delete modal (dead-field audit PR 5) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4639
• test(ui-tests): real assertions for benchmark config dropdowns + start button (dead-field audit PR 6) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4637
• chore(ui-tests): remove dead-field orphan enterKeySubmits (dead-field audit) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4648
• test(ui-tests): convert dead ctrlEnterSubmit orphan into a CI-safe keyboard-submit test by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4652
• chore(ui-tests): de-rot ctrlEnterSubmit comment + use optional chaining (review follow-up) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4656
• test(ui-tests): real assertion for flashMessagesDisplayed (dead-field audit) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4657
• test(ui-tests): assert /auth/change-password form directly, fail not skip (dead-field audit) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4662
• test(ui-tests): real assertions for crud operations (dead-field audit) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4677
• test(ui-tests): real assertions for history page (dead-field audit) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4675
• test(ui-tests): real assertions for research workflow (dead-field audit) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4676
• test(ui): AI-review follow-ups to dead-field audit tests by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4725

🐍 Python Changes

• chore: add Session type hints to remaining db_session parameters by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4575

Other Changes

• docs(test): link upstream happy-dom#2182 for the KaTeX sanitize workaround by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4438
• fix(security): pass Google Gemini API key via header (#4184) by @Donny-Guo in https://github.com/LearningCircuit/local-deep-research/pull/4452
• fix(security): suppress torch GHSA-rrmf-rvhw-rf47, document Grype DB-lag policy by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4535
• refactor(lmstudio): ensure URLs end with /v1 for LM Studio integration by @jactan77 in https://github.com/LearningCircuit/local-deep-research/pull/4532
• refactor: centralize the default search engine as DEFAULT_SEARCH_TOOL by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4547
• test(api): remove orphaned TestApiGetConfig for deleted /config endpoint by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4556
• fix(settings): fix text_separators JSON decoding, add UI validation and fallback handling by @skbs-eng in https://github.com/LearningCircuit/local-deep-research/pull/4231
• test(ui-tests): real assertions for metrics dashboard (dead-field audit) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4668
• test(ui-tests): real assertions for realtime progress (dead-field audit) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4669
• fix(security): triage XML/node CVEs + remove SQLi false positive (#7978–#7991) by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4768
• ci(pre-commit): add author identity consistency check by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4809
• style: apply ruff-format to 8 files drifted from the pinned formatter by @LearningCircuit in https://github.com/LearningCircuit/local-deep-research/pull/4813
• fix: align file backup verifier gate with UI toggle key by @skbs-eng in https://github.com/LearningCircuit/local-deep-research/pull/4803

New Contributors

• @Donny-Guo made their first contribution in https://github.com/LearningCircuit/local-deep-research/pull/4452
• @jactan77 made their first contribution in https://github.com/LearningCircuit/local-deep-research/pull/4532

Full Changelog: https://github.com/LearningCircuit/local-deep-research/compare/v1.7.0...v1.8.0