Skip to content

CHANGELOG.md

Latest commit

 

History

History
164 lines (120 loc) · 7.93 KB

CHANGELOG.md

File metadata and controls

164 lines (120 loc) · 7.93 KB
Raw

0.2.6 (2026-07-16)

🚀 Features

  • Text2SQL structured-data pipeline: upload Excel/CSV as datasets (validate → Parquet → reference doc → DATASET#), queried from chat via Data MCP (search_datasets / describe_dataset / run_sql) with per-project catalog search (18c1a90)
  • chat: per-turn model selector (Sonnet 5 / Opus 4.8 / Sonnet 4.6) with reasoning control and SSM-backed catalog; inline chart cards (render_chart: hbar/compare/timeline/donut/stacked/scatter); question cards (ask_user); typewriter streaming and stop-response (181e1bd)
  • upgrade default agent models to Sonnet 5 and Opus 4.8 (afac1ee)
  • replace DuckDuckGo with AgentCore Web Search across agents (6ca7729)
  • add chat cancellation and harden UI resource cleanup (b3234f2)

🩹 Fixes

  • prevent SQL injection in chat history query (7bac67d)
  • websocket reconnect and cleanup (#342)
  • auth: use light color scheme for managed login so heading and field labels stay visible (237a106)
  • frontend: import fromCognitoIdentityPool from credential-provider-cognito-identity to fix Vite esbuild 'fromTokenFile' export error (#331)
  • infra: apply prettier formatting to workflow-stack (f825576)

❤️ Thank You

  • 윤창헌
  • ChangHun Yoon
  • Keita
  • yunwoong
  • Zoumana Keita @keitazoumana

0.2.5 (2026-04-14)

Agent

  • Replace custom skills system with Strands built-in AgentSkills plugin (#278)
  • Upgrade strands-agents to 1.34.1
  • Add SyntaxCheckHook — pre-flight compile() check on code_interpreter calls, cancels with clean error on SyntaxError before reaching the AgentCore sandbox (#287)
  • Remove !pip install directives from docx/pptx/xlsx/chart SKILL.md files; rely on AgentCore Code Interpreter pre-installed libraries (#282)

Bug Fixes

  • Fix /chat/projects/{id}/sessions/{id} 500 error: change ChatMessage.created_at/updated_at to datetime to match DuckDB's auto-parsed timestamp type (#283)

UI

  • Add manual refresh button to the Artifacts side panel for cases when WebSocket created events are missed (#283)

0.2.4 (2026-04-02)

Knowledge Graph

  • Refactor entity extraction into dedicated Lambda (entity-extractor) with test mode for prompt tuning
  • Split analysis-finalizer into 3 parallel Lambdas: SQS sender, page description generator, entity extractor
  • Introduce core entity normalization using LLM — groups related entities loosely for better cross-page connections
  • Store core entities in LanceDB (add_graph_keywords) for cross-document keyword search
  • Add search___graph_traverse tool (qa_ids-based graph traversal, replaces graph___graph_search)
  • Add search___graph_keyword tool (keyword similarity search via LanceDB + Neptune)
  • Remove graph-mcp Lambda; merge graph tools into search-mcp
  • Optimize Neptune queries: remove entity_limit, consolidate entity loop into single UNWIND query
  • Graph search returns qa_id/qa_index for precise QA-level results
  • Graph search filters sources to only Haiku-cited segments
  • Entity extraction prompt improvements: skip visual labels, chart axes, generic terms
  • Entity normalization prompt: loose grouping with core entity creation (one entity can belong to multiple groups)

Features

  • Add graph keywords actions for lancedb-service (#248)
  • Add get_by_qa_ids action for LanceDB (#260)
  • Add rerank action and restructure search-mcp into actions/lib (#258)
  • Delete keywords by project id (#249)

Bug Fixes

  • Fix single image OCR chunk merger race condition (add chunk_index to single payload)
  • Fix backend graph rebuild API timeout (increase Lambda invoke read_timeout to 900s)
  • Fix graph-service raw_query not passing openCypher parameters
  • Fix frontend isGraphTool check for renamed graph tools
  • Fix ToolResultDetailModal segment_id parsing for qa_id format

UI

  • Add keyword display and connected entities to graph search result modal
  • Add origin page indicators (yellow nodes) in graph search visualization
  • Add clickable origin page filter in graph search modal
  • Collapsible entity list in graph search results
  • Show Analysis/Extra labels instead of QA index numbers
  • Hide NEXT edges by default in document graph view
  • Increase graph page range limit to 100
  • Graph search shows only matched entities (not all entities on found segments)
  • Deduplicate sources by page in graph results

Search Skill

  • Rewrite search skill with document search, keyword graph search, and web search paths
  • Remove internal path labels from agent responses

0.2.3 (2026-03-23)

Security

  • Bump jsdom to ^29.0.0 (#101)
  • Bump undici to >=7.24.0 (#115)
  • Bump express-rate-limit to >=8.2.2 (#104)
  • Bump @modelcontextprotocol/sdk to >=1.27.1 (#104)
  • Bump file-type to >=21.3.2 (#113)
  • Bump ajv to >=8.18.0 (#74)
  • Bump devalue to >=5.6.4 (#107)
  • Bump yauzl to >=3.2.1 (#117)
  • Bump flatted to >=3.4.0 (#118)
  • Bump svgo to >=4.0.1 (#99, #100)
  • Bump pillow to >=12.1.1 (#61)

Dependencies

  • Bump pyjwt from 2.10.1 to 2.12.0 (#211)
  • Bump hono from 4.12.4 to 4.12.7 (#199)
  • Bump pyasn1 from 0.6.2 to 0.6.3 (#216)
  • Bump aws-sdk-dynamodb in lancedb-service (#215)

Documentation

  • Add permissions docs and update FAQ (#210)

Bug Fixes

  • Fix chunk_pdf_path UnboundLocalError in finally block (#208)
  • Fix ajv override breaking eslint on Node 25 (#213)
  • Fix imported Lambda permission issue using fromFunctionAttributes with sameEnvironment

Features

  • Add Rust PaddleOCR Lambda with MNN-based CPU inference, replacing Docker container Lambda (#229)
  • Refactor OCR processor to two-Lambda architecture: Python adapter + Rust inference
  • Remove use_doc_unwarping and use_textline_orientation OCR options from UI
  • Remove entity types and cluster nodes from Neptune graph; simplify Entity ID hash to SHA256(project_id:name)
  • Replace CodeBuild-based Rust Lambda builds with cargo-lambda-cdk RustFunction construct (#220)
  • Add toka multilingual tokenizer Lambda for keyword extraction (#212)
  • Migrate lancedb-service from Python Docker Lambda to Rust Lambda with cargo-lambda-cdk (#214, #217, #218)
  • Pass language parameter to LanceDB for keyword extraction

0.2.2 (2026-03-13)

Workflow Orchestration

  • Integrate OCR, BDA, Transcribe, and WebCrawler preprocessing into Step Functions state machine with polling loops for async job tracking
  • Add real-time progress visibility for all preprocessing steps through WebSocket notifications
  • Add English comments to all Step Functions states describing purpose and branching logic

Analysis Optimization

  • Improve analysis prompt to reduce redundant processing
  • Exclude Excel (.xlsx) and CSV files from AI analysis pipeline

Bug Fixes

  • Fix webcrawler branch completing immediately without waiting for agent to finish (add DDB polling loop)
  • Fix transcribe results not being merged into segments (check use_transcribe flag instead of missing preprocess_check.status)
  • Fix reanalysis not updating language in DynamoDB workflow data
  • Fix single tilde (~) being rendered as strikethrough in markdown across all components
  • Fix shell redirect issue in Lambda layer build causing junk file creation

0.2.1 (2026-03-10)

Infrastructure updates for large-scale document processing.

Changelog

0.2.0 (2026-03-01)

Initial release of IDP Pipeline v2.