|
| 1 | +# Release Notes |
| 2 | + |
| 3 | +The purpose of this document is to provide a quick summary of all the biggest new features added in this version and provide some additional description, video, or tutorials for major features. |
| 4 | + |
| 5 | +## Brief Summary |
| 6 | + |
| 7 | +This month's release prioritized security- and performance-related tasks such as latency optimization and improved MCP toolset authentication functionality. The DIAL Admin Panel saw several enhancements such as the ability to import/export assets, and an integrated Deployment Manager component for deploying models and toolsets directly in DIAL. Several Llama and Gemini models were added to our adapters, as well as default configuration settings for Nano Banana. Feel free to watch a video showcase of some of the past month's biggest changes below: |
| 8 | + |
| 9 | +[](https://www.youtube.com/watch?v=G_-syAR2olQ) |
| 10 | + |
| 11 | +## Major Enhancements |
| 12 | + |
| 13 | +**Performance Optimization**: In addition to security and authentication enhancements for toolsets, this release included performance-related housekeeping tasks which improve load-based scalability. For example, we were able to significantly reduce our max response time (from several seconds down to 500ms) by fixing a [lock edge-case](https://github.com/epam/ai-dial-core/pull/1275), bringing both mean and median response time under 40ms, with 99th percentile remaining stable under 100 ms. Additionally, we added support for WebSocket protocol in custom routes, which improves how we handle streaming responses not just in chat (which was already supported) but also in the DIAL core. The DIAL team continues to prioritize enterprise-level stability and scalability. |
| 14 | + |
| 15 | +``` |
| 16 | +======================================================================================================================== |
| 17 | +---- Global Information -------------------------------------------------------------|---Total---|-----OK----|----KO---- |
| 18 | +> request count | 36,000 | 36,000 | - |
| 19 | +> min response time (ms) | 18 | 18 | - |
| 20 | +> max response time (ms) | 576 | 576 | - |
| 21 | +> mean response time (ms) | 38 | 38 | - |
| 22 | +> response time std deviation (ms) | 19 | 19 | - |
| 23 | +> response time 50th percentile (ms) | 37 | 37 | - |
| 24 | +> response time 75th percentile (ms) | 43 | 43 | - |
| 25 | +> response time 95th percentile (ms) | 51 | 51 | - |
| 26 | +> response time 99th percentile (ms) | 85 | 85 | - |
| 27 | +> mean throughput (rps) | 60 | 60 | - |
| 28 | +---- Response Time Distribution ---------------------------------------------------------------------------------------- |
| 29 | +> OK: t < 50 ms 34,045 (94.57%) |
| 30 | +> OK: 50 ms <= t < 150 ms 1,817 (5.05%) |
| 31 | +> OK: t >= 150 ms 138 (0.38%) |
| 32 | +> KO 0 (0%) |
| 33 | +======================================================================================================================== |
| 34 | +``` |
| 35 | + |
| 36 | +**DIAL Admin Improvements**: The DIAL Admin Panel is designed to provide a UI for system administrators to control system configuration. It now supports importing and exporting configurations of application and toolset assets, and adds an improved flow for publications processing where users can select files or toolsets to publish along with the asset itself. An integrated Deployment Manager allows administrators to deploy and host their own models, MCP servers, and interceptors. Clients often have several deployments of DIAL running simultaneously (Prod, UAT, and Dev, for example), so streamlining the process of synchronizing, maintaining, and upgrading these environments was a big priority for this release. |
| 37 | + |
| 38 | +**MCP Authentication Enhancements**: This release includes several security-related features for Toolset authentication. DIAL now supports proxying connection to MCP toolsets, which is often required to connect securely to private or firewalled MCP servers. While users could always create and connect their own MCP toolsets in the DIAL UI, we have added support for system administrators to deploy MCP toolsets requiring authentication both through DIAL Config files and through the DIAL Admin Panel. These admin-deployed toolsets can be made available to end-users based on their roles and permissions. In summation, this release heavily prioritized smooth integrating with password-protected external systems using MCP. |
| 39 | + |
| 40 | +**Adapters**: In this release, we have added support for Llama 4 Scout 17B and Llama 4 Maverick 17B, as well as Gemini 3 Pro + Pro Image (Nano Banana). Additionally, we have added some configuration parameters for vertex and bedrock adapters to control retries in http client, which will help load balancer logic handle major retry functionality. We have also added parameters to the image generation configuration for default parameters supported by Nano Banana like aspect ratio and size. |
| 41 | + |
| 42 | +## Additional Notes |
| 43 | + |
| 44 | +For full technical release notes with all bug fixes and additional features, please consult the [upgrade guide](upgrade-to-1.39.md) with all the tags for each component, as well as the DIAL documentation. |
| 45 | + |
| 46 | +* **Application schemas**: Add assistantAttachmentsInRequestSupported which allows us to build application types supporting retrieval of file content from previous responses, to improve how attachemnts are chained together. |
| 47 | +* **Multiple IDPs of same type**: DIAL always supported multiple IDPs in one instance of DIAL, such as Auth0 and Keycloak. We have added support to have multiple instances of the same IDP (ie: two different Keycloak providers) |
0 commit comments