index() now loads existing rows and applies a content diff instead of
delete-all/reinsert-all: unchanged chunks keep their rows and embeddings
(zero HNSW/GIN churn), moved chunks get a position-only UPDATE, and only
new texts are embedded, batched with the summary embedding. First index
keeps the cache-aware build_chunk_embeddings path.
Split _compute so the incremental edit path can reuse the exact same chunker
selection and embedding entry points (and their test patch targets) without
going through the doc-level cache.
Greedy multiset match on chunk text decides which rows keep their embeddings,
which texts need embedding, and which rows are deleted. No DB, no embeddings;
fully unit-tested (reuse, head insert, middle edit, deletion, duplicates,
reorder, full rewrite).
The cached payload is the indexing pipeline's embeddings (markdown is
chunked then embedded), so "embedding cache" names the expensive output
directly and removes the "index" ambiguity (DB index vs vector index vs
indexing phase). Renames the service, settings, eligibility, eviction
task, metrics, config flags (INDEX_CACHE_* -> EMBEDDING_CACHE_*), object
prefix, and the table (index_cache_embedding_sets -> embedding_cache_sets)
with its constraint and indexes. Migration 161 renamed accordingly.
- Introduced LLMErrorCategory and adapt_llm_exception to normalize LLM exceptions.
- Updated llm_retryable_message and llm_permanent_message to utilize the new adaptation logic.
- Enhanced classify_stream_exception to classify provider errors and return user-friendly messages.
- Added tests for error classification and adaptation to ensure robustness.
- Updated frontend error handling to display appropriate messages based on new classifications.
Document_chunker currently splits Markdown tables mid-row when the table is
larger than a single chunk window, producing garbled rows that are useless
for RAG retrieval (issue #1334).
Changes:
- document_chunker.py: add chunk_text_hybrid() that detects Markdown table
blocks with a regex, emits each table as an indivisible single chunk, and
feeds the surrounding prose through the normal chunk_text() chunker.
- indexing_pipeline_service.py: route normal (non-code) documents through
chunk_text_hybrid instead of chunk_text so tables are protected by default.
Fixes#1334
- Added performance logging to the `index_batch_parallel` method, capturing metrics for document indexing duration and concurrency.
- Introduced timing measurements for both the overall indexing process and the parallel document gathering phase, improving observability of the indexing workflow.
- Updated logging statements to provide detailed insights into the number of documents processed, indexed, and failed during the indexing operation.
- Refactored Google Calendar and Gmail indexers to utilize the new `index_batch_parallel` method for concurrent document indexing, enhancing performance.
- Updated the indexing logic to replace serial processing with parallel execution, allowing for improved efficiency in handling multiple documents.
- Adjusted logging and error handling to accommodate the new parallel processing approach, ensuring robust operation during indexing.
- Enhanced unit tests to validate the functionality of the parallel indexing method and its integration with existing workflows.
- Added `index_batch_parallel` method to enable concurrent indexing of documents with bounded concurrency, improving performance and efficiency.
- Refactored existing indexing logic to utilize `asyncio.to_thread` for non-blocking execution of embedding and chunking functions.
- Introduced unit tests to validate the functionality of the new parallel indexing method, ensuring robustness and error handling during document processing.
- Added `download_and_extract_content` function to extract content from Google Drive files as markdown.
- Updated Google Drive indexer to utilize the new content extraction method.
- Implemented document migration logic to update legacy Composio document types to their native Google types.
- Introduced identifier hashing for stable document identification.
- Improved file pre-filtering to handle unchanged and rename-only files efficiently.
- Added endpoint to list agent tools with metadata, excluding hidden tools.
- Updated NewChatRequest and RegenerateRequest schemas to include disabled tools.
- Integrated disabled tools management in the NewChatPage and Composer components.
- Improved tool instructions and visibility in the system prompt.
- Refactored tool registration to support hidden tools and default enabled states.
- Enhanced document chunk creation to handle strict zip behavior.
- Cleaned up imports and formatting across various files for consistency.
- improved search_knowledgebase_tool
- Added new endpoint to batch-fetch comments for multiple messages, reducing the number of API calls.
- Introduced CommentBatchRequest and CommentBatchResponse schemas for handling batch requests and responses.
- Updated chat_comments_service to validate message existence and permissions before fetching comments.
- Enhanced frontend with useBatchCommentsPreload hook to optimize comment loading for assistant messages.
- Introduced RequestPerfMiddleware to log request performance metrics, including slow request thresholds.
- Updated various services and retrievers to utilize the new performance logging utility for better tracking of execution times.
- Enhanced existing methods with detailed performance logs for operations such as embedding, searching, and indexing.
- Removed deprecated logging setup in stream_new_chat and replaced it with the new performance logger.
- Replaced direct embedding calls with a utility function across various components to streamline embedding logic.
- Added enable_summary flag to several models and routes to control summary generation behavior.
- Introduced slow callback logging in FastAPI to identify blocking calls.
- Added performance logging for agent creation and tool loading processes.
- Implemented caching for MCP tools to reduce redundant server calls.
- Enhanced sandbox management with in-process caching for improved efficiency.
- Refactored several functions for better readability and performance tracking.
- Updated tests to ensure proper functionality of new features and optimizations.
