v0.8.5

Release Notes
This release focuses on embedding model support, and improved token usage tracking.

Embeddding Model

• User can select variety of embedding models to generate vector embeddings for your data. This can be configured from the frontend in Graph Settings > Processing Configuration > Select Embedding Model.
• Supported model providers include OpenAI, Gemini, Amazon Titan, and Sentence Transformers.
• Your selected embedding model is saved to your user profile when TRACK_USER_USAGE is enabled.

Token usage tracking issues fixed

• Null email in for some users leading to errors of exhausted token limit #1463
• Modular approach to disable token tracking error by env variable LIMIT_TOKEN_USAGE_PER_USER
• Add a team contact option when token limit is exhausted

v0.8.4

Release Notes

This release focuses on dependency upgrades, model updates, cleaner architecture, and improved usage tracking.

Dependencies

• Updated all frontend and backend dependencies to their latest stable versions.

Authentication

• Login functionality is now enabled for the production-deployed environment.

LLM Model Updates

• Added and upgraded to the latest supported LLMs, including Gemini 2.5, OpenAI GPT-5.x, Anthropic Claude 4.5, Bedrock Nova variants, Groq, Fireworks models, and others.

• Gemini 2.5 flash is default model

Codebase Improvements

• Refactored codebase to remove unnecessary logic.
• Introduced Dependency Injection and Reduced API function parameters.
• Streamlined API flows with uniform input parameters for better consistency and maintainability.

Documentation

• Updated README and refreshed both backend and frontend documentation.
• Updated example env for better understanding of what are optional and mandatory envs required.

Token Usage Tracking

• Token usage is now tracked per user for extract, chatbot, and post-processing APIs.
• Non-Neo4j users receive a daily limit of 250,000 tokens and a monthly limit of 1,000,000 tokens.
• Limits can be reset via cron jobs.
• Repository users can enable tracking using TRACK_USER_USAGE=true.
• Current usage is visible in the UI under the logged-in user section.

v0.8.3

🚀 Features
• Added Data Importer support – import graph models from Neo4j Console Preview into LLM Graph Builder. (#1301 )
• Claude 4 Sonnet (Anthropic) now available in production. (#1299 )
🛠️ Updates
• Backend dependencies upgraded to latest versions. (#1321 )
🐞 Fixes
• Protobuf file duplicate error resolved (#1323 )
• Resolved issue with missing YouTube transcripts. (#1321)

v0.8.2

What's Changed

Bug fixes

• Connecting to different db due to race condition #1283

• chatollama not working #1286

• Tuple schema validation added #1289 #1290

Full Changelog: v0.8.1...v0.8.2

v0.8.1

What's Changed

#1246 minimize the docker file size using torch CPU version
#1275 Dropdown updates with all required source, type, target values from user selection of Predefined Schema, Load Existing schema, Schema from text.
#1248 Validation for tuple schema to avoid multiple value inputs in source, type and target

Full Changelog: v0.8...v0.8.1

v0.8

🚀 New Features

• Schema Visualization Tool: Introduced a powerful schema visualization interface with multiple options:

• Generate from Text: Users can now input plain text to extract schema in triplet format (Source → Relationship → Target).

• Load from Database: Visualize existing schema directly from the database in triplet format.

• Predefined Schemas: Choose from a library of domain-specific schemas (e.g., Retail, Healthcare) to jumpstart your project.

• User-Defined Schema: Users can define their own schema by selecting from a dropdown or typing directly in triplet format (Source → Relationship → Target).
  #1235 #1230.

• Legacy Format Migration: Easily convert previously stored schema formats to the new triplet-based format for consistency and improved visualization.

⚙️ Enhancements

• LLM Upgrades (Production): Replaced Gemini 1.5 Flash with the more powerful Gemini 2.0 Flash for enhanced performance in production workflows. [#1233]
• LLM Additions (Development): Integrated support for LLaMA4 Maverick, LLaMA4 Scout, and OpenAI GPT-4.1 in the development environment.[#1233]

🛠 Bug Fixes & Improvements

• File Upload Status: Fixed an issue where failed file uploads were not correctly reflected in the database status.[#1222]

• Accurate Node & Relationship Count: Updated logic to correctly reflect node and relationship counts during the extraction process. [#1191]

• Deadlock Handling: Resolved transient deadlock errors (Neo.TransientError.Transaction.DeadlockDetected) by implementing up to 3 automatic retries for affected Cypher queries. [#1187]

📦 Dependency Updates

• Upgraded all backend packages to their latest versions for improved compatibility and performance. [#1189]

v0.7.2 Bug Fix Release

🔒 Security Enhancements
• Changed API Method for Security: The sourcelist API has been changed from GET to POST to prevent password exposure. (#1102)
🛠 Bug Fixes & Improvements
• Backend Fixes ((#1097) :
Fixed list index out of range error.
Resolved GCS file not found issue.
Additional data fetching from related links in wikipedia #1151
Top entities not found for some modes #1150 #1154

Extraction Process Fixes:
• Docx File Extraction Issue: Resolved missing pandoc dependency by adding pypandoc-binary.
(#1124)
• UnicodeDecodeError Fix: Addressed gb2312 encoding issue for certain text files.
(#1126)
• GraphDB Access Bug: Fixed UnboundLocalError during extraction.
(#1129)
• Connection Issues Resolved:
• Fixed “connection not there” error when accessing data resources (#1131).

📝 Code & Configuration Cleanup
• Removed example.env from the root folder to prevent confusion.
(#1099)
• Sanitized Additional Instructions: Prevented issues due to curly brace formatting and potential prompt injection.
(#1130)

🎨 UI Fixes & Enhancements
• Schema button disabled when DB is not connected.
• Data sources hidden on small screens if DB is not connected.
• Added an information message when visualizing graphs with more than 50 chunks.
(#1097)

v0.7.1

What's Changed

New features

• schema visualization -#1035

• Package updates : langchain neo4j and other langchain packages - #1048

• Cancelling the API's on Unmounting phase - #1068

Bug fixes

• UI fixes - #1091

• Backend connection config fixed - #1060

• Removed chatbot related entities from get existing schema - #1061

• Resolved web url sources extraction failing for some of the urls due to leading and trailing spaces in document names. - #1064

• Support added for new models - #1069

1. gemini 2.0 flash
2. GPT o3 mini (In dev)
3. deepseek r1 & v3 (In dev)

Full Changelog: v0.7...v0.7.1

v0.7

Enhanced User Experience & Insights

• File Expiration Alert:
Users are now notified when local files have expired, ensuring timely action. #953
• Fallback for Blank Wikipedia Titles:
Titles are now assigned based on the URL when metadata titles are unavailable, improving data clarity. #982
• Additional Instructions:
A new tab has been introduced to allow users to provide specific instructions for entity extraction, such as focusing on key topics. #1013, #1014

Improved Knowledge Graph Construction & Retrieval

• Limit Processed Chunks:
Introduced restrictions on the number of chunks created and processed, enhancing performance and scalability. #1000
• Graph Consolidation Changes:
New logic consolidates large graph schemas into fewer, more relevant node labels and relationship types, improving graph quality. #1013, #1014
• New Models Integration:
Integrated Amazon Nova models (Micro, Lite, Pro v1) for graph generation and chatbot Q&A, along with Titan embedding model support. #1006
• Effective Search Ratio Parameter:
Introduced the effective_search_ratio parameter to improve query accuracy by expanding the Lucene index candidate pool. Configurable via backend environment variables. #981
• Custom Error Handling for APIs:
Introduced LLMGraphBuilderException to handle user-facing errors in extract and url_scan APIs, ensuring better feedback. #989

Code Refactoring

• Code Cleanup:
Removed unused libraries and commented-out code for improved maintainability. #973
• Post-Processing Updates:
Removed the isSchema check for graphType in post-processing jobs, streamlining logic.
• Documentation Updates:
Updated README.md and front-end documentation for better clarity. #974
• Driver Optimization:
Ensured the driver is properly closed after fetching chunk text details, avoiding resource leaks. #938

Bug Fixes

• Metric Table Fixes:
Resolved various UI issues with the Metric table, ensuring smoother performance. #921
• Database Connection Issues:
Fixed issues when switching database instances, including:
• Refreshing the chatbot screen when switching.
• Addressing atob console errors and duplicate key warnings.
• Removed strip() function from document names. #966
• Frontend Improvements:
Addressed .env model format errors, replacing unclear messages with more descriptive error messages. #946
• Schema Validation Fixes:
Handled EquivalentSchemaRuleAlreadyExist errors with improved validation. #949
• Logger Error Fixes:
Resolved issues with logger errors caused by JSON parsing. #994

v0.6

Enhanced User Experience & Insights:

• Chunk Details on UI - Users can now view the text chunks extracted from each document source directly in the UI. These chunks power the knowledge graph and enable precise information retrieval using various strategies. #850
• Node & Relationship Counts - The UI now provides a detailed breakdown of chunk, entity, and community node counts for each document source, offering deeper insights into the knowledge graph structure. #881, #890
• Enhanced Security for Private Instances - Backend services no longer expose Neo4j credentials to the frontend, improving security for privately hosted deployments. #896
• Dedicated Chat Interface - A streamlined URL is now available for accessing the chat functionality directly. #870

Improved Knowledge Graph Construction & Retrieval

• Connectedness for Extraction - Enhanced logic for chunk extraction improves the connectedness and overall quality of the knowledge graph. #852
• Single Chunk Extraction - Switched to a single chunk extraction method in asynchronus way, streamlining the process of KG extration. #901
• New Neo4j Langchain Package - Introduced a new top-level Neo4j Langchain package and upgraded other packages to their latest versions, simplifying integration and usage. #891

Advanced Evaluation & Metrics

• Expanded Chatbot Evaluation Metrics - Introduced new metrics for evaluating chatbot answers, including ROUGE score, semantic similarity score, and context entity recall score. #855

Bug Fixes

• Check for file existence in GCS bucket, Updated condition for Reprocess from last processed position. #917
• Duplication of node query update and dynamic vector index dimension based on embedding model in post processing. #929
• Community creation check from UI. #916
• Database icon to distinguish between different Neo4j graph database instances. #924