This release is a Compatibility Level 2 update. For details, please refer to the V3.00.6 Release Notes.
Features marked [V3.0] are available exclusively in DolphinDB 3.00.6.
As enterprises race to turn AI from pilot projects into production systems, the real bottleneck is often integration — connecting data, compute, and workflows into something secure and reusable. With DolphinDB V3.00.6 & V2.00.19, we’re tackling this head-on.
Central to this release is DolphinX, an enterprise-grade platform that deeply integrates DolphinDB’s native capabilities with an Agent platform — turning data, computing, and business logic into secure, reusable, and governable AI agent capabilities with minimal effort.
This release also strengthens core capabilities across FeatureDB, FICC, JIT compilation, dynamic script optimization, high-availability stream computing, and system operations — spanning the full workflow from low-latency feature storage and complex derivatives pricing to high-performance script execution and enterprise-grade governance.
1. DolphinX: An Enterprise-Grade AI Development and Governance Platform
Available in V3.0
As demand grows for natural language data querying, intelligent analytics, and automated workflows, the traditional “database + external AI framework” approach is increasingly challenged by concerns around security, compatibility, operational complexity, and enterprise governance. DolphinX, an enterprise-grade platform for AI agent development and governance, enables organizations to transform their data, computing, and business capabilities into secure, governable, and reusable AI agent services with a single integration, empowering faster business innovation and accelerating AI deployment.
Natural Language-Driven Data Workflows
With the built-in Coding Agent, users can interact with DolphinDB’s core capabilities using natural language, eliminating the need to learn Dlang. From script generation and debugging to SQL query analysis, data import/export, stream processing, and real-time engine development, Coding Agent makes advanced data capabilities accessible to a much broader range of business users.
Out-of-the-Box AI Toolkit with Flexible Extensibility
DolphinX works out of the box with zero configuration required. Just log in through the web interface to get started. It also includes a web-based management console for configuring prompts, models, and MCP servers, and attaching Skills to build dedicated AI agents.
DolphinX comes with a rich set of built-in Skills covering coding standards, data import, financial services, stream processing, operations and maintenance, machine learning, data visualization, testing, and more (see the table below).
For organizations requiring deeper customization, DolphinX exposes APIs that enable seamless integration of its AI capabilities into existing enterprise systems.
Press enter or click to view image in full size
Eliminate External Dependencies and Build an Enterprise-Grade Intelligent Closed Loop
AI agent capabilities are built directly into the DolphinDB Server, eliminating the need for third-party orchestration frameworks. This enables a complete, end-to-end workflow — from understanding and analysis to execution — entirely within a single system. For scripts generated by AI models but not yet reviewed by users, DolphinX parses and validates them before execution, blocking high-risk operations such as deletion and cleanup. When invoking DolphinDB capabilities, AI agents inherit the current user’s permissions, eliminating the need for a separate access control system. This significantly reduces the security risks, compatibility issues, and operational complexity associated with external dependencies, making enterprise AI applications more secure, reliable, and efficient.
To further enhance AI’s understanding of database environments, DolphinDB has also expanded its semantic metadata framework for database objects. Building on existing support for descriptions of tables, table columns, and user-defined functions, this release introduces description support for both the Catalog and Schema levels. In addition, the maximum length of table column descriptions has been increased from 256 bytes to 1,024 bytes. By providing richer representations of database hierarchy and business semantics, AI agents can more accurately understand data assets, generate scripts, and execute analytical tasks, resulting in a more reliable enterprise-grade intelligent closed loop.
Enterprise-Grade Agent Development and Governance Foundation
DolphinX provides a common capability layer for building and governing Agents, including session management, Memory, Skills, context assembly, LLM provider management, and permission auditing. This helps enterprises avoid rebuilding these capabilities from scratch for every Agent, while enabling unified governance and operations.It supports both short-term and long-term memory, and provides a built-in RAG knowledge system based on DolphinDB’s TextDB and VectorDB. It also supports unified integration with multiple LLM providers, including fallback switching, token budget management, and local or private cloud deployment, balancing flexible extensibility, data security, and autonomous control.
Designed for scenarios such as natural language interaction, intelligent analytics, and automated processing, DolphinX deeply integrates AI capabilities with DolphinDB to provide an integrated product ecosystem that combines out-of-the-box usability with enterprise-level extensibility. It enables organizations to build intelligent data applications with lower barriers to entry and reduced system integration costs.
2. FeatureDB: A Low-Latency Feature Storage Engine for AI and Machine Learning Applications
Available in V3.0
In AI and machine learning training and online inference scenarios, feature data access follows different patterns than general analytical queries with high-frequency reads by column, window, and timestamp. Traditional online feature stores, which are primarily designed for row-based access, focus more on full-row read/write operations and key-value retrieval. As a result, they are not fully optimized for these access patterns and can become performance bottlenecks in scenarios involving wide feature tables, frequent reads, and low-latency services.
FeatureDB is purpose-built to address exactly this gap. Running within the DolphinDB environment, it offers a lightweight, high-performance foundation for feature serving pipelines in AI and machine learning applications.
Note: The FeatureDB introduced in this release currently supports in-memory mode only. Persistent storage will be available in future releases.
Microsecond-Level Random Reads to Alleviate Feature Access Bottlenecks
FeatureDB delivers microsecond-level random reads, with response times typically ranging from a few micro-seconds to tens of microseconds. This significantly reduces the impact of feature access latency on both training throughput and inference speed, especially in large-scale, high-frequency access scenarios.
Native Support for Machine Learning Data Types and Flexible Wide Table Design
FeatureDB natively supports commonly used floating-point data types in machine learning, including FLOAT8, FLOAT16, FLOAT32, and FLOAT64, covering different tradeoffs between precision and storage efficiency. It also supports wide tables containing hundreds of feature columns and provides simple syntax for quickly defining continuous feature columns, reducing the management complexity of large-scale feature engineering.
Seamless Python Ecosystem Integration, Simplifying Workflows from Import to Retrieval
On the upstream side, FeatureDB integrates with feature computation tools such as Spark and PyArrow through the standard Parquet format, enabling one-click batch import. On the downstream side, its Python SDK directly outputs data that can be consumed by mainstream ecosystem tools such as NumPy and PyTorch, covering the complete pipeline from storage to model training and inference. By leveraging industry-standard tools and formats throughout the workflow, FeatureDB effectively reduces system integration and maintenance costs.
Press enter or click to view image in full size
The introduction of FeatureDB marks an extension of DolphinDB’s capabilities in AI and machine learning scenarios, providing users with a new infrastructure option for building high-performance feature data services.
3. Building a Complete Pricing Loop: Comprehensive FICC Capability Enhancements
Available in V3.0
The new release continues to enhance DolphinDB’s FICC pricing capability framework by introducing a wide range of pricing functions across fixed income, interest rates, credit, foreign exchange, and equity OTC derivatives. New support covers agreements, government bond futures, FRAs, Caps/Floors, Swaptions, CDS, NDFs, digital options, and range accrual options.
In addition, the release introduces new foundational tools such as Copula functions, credit curves, volatility surfaces/cubes, date generation, and schedule construction. These enhancements further strengthen DolphinDB’s cross-asset pricing and risk analysis capabilities, enabling a more complete and integrated pricing workflow.
Press enter or click to view image in full size
4. Enhanced JIT Capabilities with Broader Language Support and Greater Stability
Available in V3.0
As enterprises increasingly demand real-time execution of complex business logic, high-performance script acceleration, and event-driven processing, Just-In-Time (JIT) compilation has become a critical foundation for executing core computational workloads. In production environments, users expect more than successful compilation — they also require comprehensive support for complex language features, rich object and container operations, and consistent performance under long-running workloads, complex event streams, and high-concurrency scenarios.
DolphinDB has comprehensively upgraded its JIT engine in the new release, expanding language support while significantly improving runtime stability. These enhancements allow users to migrate more business logic into the high-performance execution path with minimal code refactoring.
Broader Language Support for Complex Scripts
The new release adds JIT support for a wider range of language features, including partial application, anonymous functions, lambda expressions, closures, default parameters, and keyword arguments. As a result, many scripting patterns that previously relied on the interpreter can now be compiled directly without rewriting code specifically for JIT compatibility. This significantly improves scripting flexibility for scenarios involving reusable abstractions, complex computational workflows, and sophisticated data structures.
Enhanced Type System and Container Support
This release introduces explicit type annotations as a major enhancement to the JIT type system. Developers can declare member types directly within class definitions using the following syntax:
class MyClass {
field1 :: STRING
field2 :: INT
field3 :: DOUBLE VECTOR
field4 :: DICTIONARY[STRING, STRING]
}Type annotations support scalar types, user-defined classes, vectors, ANY, and recursively nested dictionaries. The JIT compiler generates type-specialized machine code directly from these declarations, eliminating runtime type inference and significantly improving performance when accessing complex objects and nested data structures.
Container operations have also been substantially expanded. The JIT runtime now supports reading, assignment, deletion, clearing, indexing, and concurrent access across these data types. In addition, Matrix support and non-scalar indexing have been introduced, while numerous container-related crashes, memory issues, and incorrect computation results have been resolved.
Improved Compatibility for CEP
JIT support for classes has been further enhanced to better serve Complex Event Processing (CEP) applications. The new release supports event classes, monitor classes, and helper classes used in CEP, while improving advanced class features such as dictionary/vector member access, nested classes, inter-method calls, and callback function passing. These enhancements improve both execution efficiency and runtime stability in CEP-related workloads.
Greater Stability and Production Readiness
Beyond language enhancements, this release significantly improves the engineering robustness of the JIT engine. Numerous stability issues involving complex data processing, nested assignments, loop execution, classes, and tuples have been resolved, along with multiple memory leaks, persistent memory growth, and abnormal resource utilization issues. Together, these improvements reduce runtime failures and make JIT considerably more reliable for complex business logic running under production-scale data workloads.
5. Upgraded Dynamic Script Optimization for Faster Nested Functions and Loop Execution
In quantitative analytics, real-time risk management, and low-latency trading, users frequently write complex scripts containing loops, conditional branches, and function calls. Because these workloads are often difficult to fully vectorize, interpreter overhead can become a significant performance bottleneck.
DolphinDB provides Dynamic Script Optimization, which automatically optimizes interpreted scripts at runtime without requiring any code modifications. Once enabled, it significantly accelerates scripts containing loops, conditional branches, and function calls. In function-intensive workloads, performance improvements of several times have been observed.
In the new release, this capability has been further refined by changing the runtime optimization scope from process-level to session-level. Configuration parameters continue to define the default behavior, while runtime APIs now affect only the current session without impacting other sessions within the same process. This improvement provides much greater flexibility in multi-user and multi-task environments. For example, users can enable optimization in one session for performance testing while keeping another session unchanged for functional validation, making the feature more practical for production deployment.
Dynamic Script Optimization is particularly effective for the following scenarios:
- Deeply nested function calls, such as strategy logic decomposed into multiple reusable functions with frequent invocation.
- Complex iterative computations, including tick-by-tick or bar-by-bar processing, element-wise window calculations, and recursive state updates.
- High-frequency conditional branching, such as signal generation, chained risk control rules, and sequential trading condition filtering.
- Business logic that cannot be easily vectorized, including state-dependent, path-dependent, or multi-stage decision workflows.
For these workloads, Dynamic Script Optimization significantly reduces interpreter overhead and minimizes the performance cost of loops and function calls. Benchmark results show performance improvements of 63.9%–88.8% for function-intensive scripts, up to 8.93× acceleration for deeply nested function calls, and more than 30% performance gains for representative financial factor calculations such as VWAP.
6. Enhanced High Availability for Streaming Data
The new release further strengthens the high availability (HA) capabilities for streaming data, covering both HA stream tables and HA MVCC tables to improve production reliability.
- For stream tables, data subscription and delivery are now more stable, subscription management is more reliable, and services recover faster after node status changes. These enhancements also extend to the ORCA streaming platform, improving stream graph continuity and overall data reliability.
- For MVCC tables, HaMvccTable further enhances the stability of data reading, writing, and synchronization, improves resource management, and ensures data consistency across multiple nodes. With continued optimization, HaMvccTable is now more production-ready and provides stronger protection for mission-critical data.
7. Improvements to Core Capabilities
This release also delivers enhancements across the database kernel and data analytics engine.
Database
Core database capabilities have been improved across distributed scheduling, storage integration, data lifecycle management, and write control. The new release delivers better task scheduling balance in multi-replica clusters, simplified Amazon S3 integration through the AWS SDK default credential chain, enhanced monthly partition retention and cold storage migration, and more precise write controls to prevent unintended NULL values during complex update operations.
Analytical Functions
The function library has been expanded for quantitative research and real-time analytics. New capabilities include time-based weighted calculations, linear filtering, variable-length sliding windows, and rolling drawdown calculations. Custom aggregation and regression options have also been extended, while text import and JSON serialization have been improved for better interoperability with external systems.
Core Language
The core language continues to evolve with improvements to type definitions, class support, control flow, metaprogramming, SQL semantics, and function metadata. New SQL features include union join (uj), ROLLUP, group-level pagination with context by, enhanced type conversion, and improved identifier compatibility. Function annotations have also been enriched with @desc, parameter types, and return type declarations.
Stream Processing
The streaming framework has been further strengthened with more flexible trigger mechanisms in the time-series engine, refined snapshot generation across multiple time ranges, access control identifiers for stream table queries, and improved management of dynamic metrics in ORCA stream graphs, delivering greater stability for complex streaming workloads.
Plugins & API
Plugin installation and Python data upload have been streamlined, improving plugin deployment in restricted network environments. Python integration with the Apache Arrow ecosystem has also been enhanced, enabling smoother interoperability with tools such as Polars.
8. Enhanced System Operations
The new release strengthens observability, resource management, and operational efficiency for production environments.
Better Observability
System monitoring has been improved with richer metadata and status visibility. Administrators can now query historical background tasks based on permissions, while stream tables, functions, shared variables, and in-memory tables expose additional status information. Automatic DDL generation through getDatabaseDDL and getDBTableDDL also simplifies database migration and schema replication.
Stronger Security and Resource Control
AES-256-CBC encryption has been added for secure string encryption and decryption. License management now supports hardware dongle verification and real-time license status monitoring. User-level task concurrency limits help prevent resource contention, while Catalog and Schema objects now support descriptions for easier management in multi-tenant environments. (Some features are available only in V3.00 and later.)
9. Performance and User Experience Improvements
This release delivers comprehensive improvements in data processing performance, system stability, and usability. Enhancements to query parsing, error reporting, storage scheduling, cache warm-up, transaction management, and Leader stability improve performance, observability, and reliability while reducing resource consumption in storage-compute separation, hot/cold tiering, and large-scale data access scenarios.
At the storage layer, TSDB introduces the new volumeType configuration for storage volumes. With SSD optimization enabled, level file access paths have been optimized to reduce storage latency and improve throughput for high-frequency read workloads. Benchmark results show an average throughput improvement of approximately 13% for multi-level group aggregation, with gains of up to 35% in specific scenarios. CPU-intensive read workloads achieve throughput improvements of up to 109%.
Usability has also been enhanced with improvements to object display, timestamp handling, dynamic metric management, field parsing, license queries, matrix indexing, and other commonly used data processing capabilities, providing greater flexibility, ease of use, and compatibility.
10. Roadmap
Storage & Data Ingestion
- Introduce persistent mode for FeatureDB to improve the reliability and reusability of feature data.
- Enhance the read performance of external tables (e.g., Amazon S3) to improve data ingestion efficiency in lakehouse architectures.
Financial Derivatives Pricing
Expand support for pricing complex financial derivatives across three major asset classes:
- Fixed Income: Embedded-option bonds
- Credit: CLN, CRMW, and CRMA
- FX: Asian, Barrier, Touch, and Quanto options
Further enhance core pricing models — including LV, SV, SLV, PDE, and Monte Carlo methods, and optimize pricing engine compatibility to provide a unified infrastructure for multi-asset derivatives pricing.
AI Agent Ecosystem
Launch a suite of domain-specific AI agents for different use cases:
- Database Operations Agent
- Plugin Development Agent
- Machine Learning & Data Analytics Agent
- Quantitative Research Agent
- Dlang Programming Assistant with long-context conversation support.
- Release an AI agent evaluation framework X-Lab for measuring and validating agent capabilities.
Query Engine Optimization
Continue enhancing the SQL engine by improving standards compliance while significantly boosting execution performance for complex multi-table join workloads.
Download DolphinDB V3.00.6 & V2.00.19 now.