Scalytics Connect 3.5

Scalytics Connect

The June 2025 update for Scalytics Connect delivers critical stability improvements to the Deep Search agent and the core API. This release focuses on resolving complex, intermittent issues related to network proxies, context window management, and API configuration, resulting in a more reliable and resilient platform for all users.

‍

Key Highlights for June 2025: Stability & Bug Fixes

‍

• High-Performance vLLM Serving Engine:
  • Torch Model Support: All local, non-embedding models now run on the vLLM inference engine, providing significant performance gains, lower latency, and higher throughput for concurrent requests.
• Enhanced AI Capabilities:
  • Advanced Stream Processing: The vLLM integration enables more robust and efficient real-time stream processing, making chat interactions and agentic workflows smoother and more reliable.
  • Deep Search Optimization: The Deep Search agent now leverages the power of vLLM for its planning, analysis, and synthesis steps, resulting in faster and more complex research capabilities.
  • Foundation for Image Generation: The integration of vLLM lays the groundwork for supporting high-performance image generation models in future releases.
• Core System & API Improvements:
  • Primary Model Detection: The Scalytics API (/v1/chat/completions) now correctly identifies the active vLLM model as the default, resolving a critical configuration bug and ensuring seamless API access without manual model selection.
  • UI Enhancements: The hardware monitoring dashboard has been updated to accurately reflect the model assigned to each GPU, providing administrators with a clear view of resource utilization in the new vLLM environment.
• Dynamic Context Management: The system now intelligently detects and utilizes the full context window of the active model, eliminating previous errors related to incorrect context size limits.

‍

Scalytics Federated AI MCP features:

• AI-based optimizer automatically picks an optimal configuration of class data processing frameworks, such as Java Streams or Apache Spark, to run applications on.
• Scalytics Core performs the program execution. It abstracts the different platform specific APIs and coordinates inter-platform communication.
• Applications can run on multiple data processing platforms, without changing the native code of the underlying platforms.
• Federated data processing: In-situ processing in different sites without moving raw data outside their origin.
• Build and execute cross platform machine learning pipelines in a unified way.
• NEW: Federated Machine Learning
  • Federated analytics by integrating multiple platforms across silos
  • Developers: Train ML models using federated learning in a platform agnostic way
• NEW: Supporting unsupervised learning (e.g., using K-means) and Stochastic Gradient Decent optimization technique for Federated Learning across supported data platforms
• NEW: Auditing compliance (who accessed what when) and training audits (basic)

‍

Data sources:

• PostgresSQL
• Columnar Data Files (e.g., CSV, Iceberg, Parquet, ORC)
• SQlite (e.g. Mobiles, Embedded)
• Local file systems
• Distributed file systems (e.g., HDFS, S3)
• Apache Kafka
• NEW: Remote files over http(s)
• NEW: JDBC based data sources

‍

Data Processing Platforms:

• Java 8 Streams
• Apache Spark / DataBricks
• Postgres
• SQLite
• Apache Flink / Confluent, Decodable
• NEW: Apache Kafka
• NEW: Tensorflow
• NEW: JDBC based platforms

‍

Programming APIs

• Java
• Scala
• Basic SQL
• New: Python (limited support)

‍

Runtime

• NEW: Actor-based runtime for building federated applications

‍

This release represents our ongoing commitment to delivering a secure, efficient, and privacy-focused AI platform for enterprise environments. We continue to enhance Scalytics Connect based on customer feedback and emerging industry requirements, ensuring our solution remains at the forefront of enterprise AI technology.

‍