~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.
Python academiaanthropicarxivbravedeep-researchencryptionhome-automationhomeserverlocallocal-deep-researchlocal-llmmistralollamaopenaipubmedresearchresearch-toolretrieval-augmented-generationsearxngself-hosted
docs(progress): cross-reference _REPORT_PHASES emitters (#4481)
LearningCircuitd5ccbddbc5f6
diff --git a/src/local_deep_research/web/services/research_service.py b/src/local_deep_research/web/services/research_service.pyindex 0dc8eedcde..7e322875a7 100644--- a/src/local_deep_research/web/services/research_service.py+++ b/src/local_deep_research/web/services/research_service.py@@ -48,6 +48,9 @@ _global_research_semaphore = threading.Semaphore(_MAX_GLOBAL_CONCURRENT)_DETAILED_SEARCH_PROGRESS_CAP = 8 # Search/output phases capped here_DETAILED_REPORT_PROGRESS_START = 10 # Report generation starts here_DETAILED_REPORT_PROGRESS_END = 100 # Report generation ends here+# Phases that belong to the report-generation stage. "report_generation" is+# emitted in this file; the other four are emitted by report_generator.py. If+# you add or rename a phase, update both this set and the emitter._REPORT_PHASES = frozenset({"report_generation",@@ -58,18 +61,6 @@ _REPORT_PHASES = frozenset(})-# Socket.IO emission throttling: minimum interval between progress emissions per research-_EMIT_THROTTLE_SECONDS = 0.2 # 200ms-_EMIT_TTL_SECONDS = 3600 # 1 hour — evict stale entries from orphaned research--# Cap on the partial-content buffer kept server-side so chat-mode termination-# can persist whatever was already streamed. Bounded to keep memory predictable-# under pathologically long answers (typical answers are a few KB).-_MAX_PARTIAL_BUFFER_BYTES = 256 * 1024 # 256 KB-_emit_cleanup_counter = 0-_last_emit_times: dict[str, float] = {}-_last_emit_lock = threading.Lock()-# Phases that produce user-visible step messages in chat mode.# "complete" is excluded — it fires AFTER the response message is written,# which would create a step with a higher sequence_number than the response.@@ -89,6 +80,18 @@ _STEP_PHASES = frozenset(})+# Socket.IO emission throttling: minimum interval between progress emissions per research+_EMIT_THROTTLE_SECONDS = 0.2 # 200ms+_EMIT_TTL_SECONDS = 3600 # 1 hour — evict stale entries from orphaned research++# Cap on the partial-content buffer kept server-side so chat-mode termination+# can persist whatever was already streamed. Bounded to keep memory predictable+# under pathologically long answers (typical answers are a few KB).+_MAX_PARTIAL_BUFFER_BYTES = 256 * 1024 # 256 KB+_emit_cleanup_counter = 0+_last_emit_times: dict[str, float] = {}+_last_emit_lock = threading.Lock()+def _chat_step_decision(phase: str | None,