Release 7.9.0 - ConvyAI
What's new
-
New ConvyAdmin home - The page shown right after login now presents the four ConvyAI modules (Bot Studio, Agentic Orchestrator, Contact Center, AVA) as cards with direct entry points, plus quick links to the configuration shared across modules. Modules not enabled on your tenant appear as informational cards. See Getting Started for details.
-
Online documentation from the ConvyAdmin header - A help icon in the header opens this documentation in a side panel without leaving the page you are working on. The content can also be opened in a separate browser tab. See Getting Started for details.
-
Reasoning effort for OpenAI and Azure OpenAI - A new Reasoning effort field in the advanced parameters of OpenAI and Azure OpenAI AI Models lets you choose the quality, latency and cost trade-off of latest-generation reasoning models. The value is configured once and applies to Knowledge Base and AI Agent nodes alike. See OpenAI and Azure OpenAI for details.
-
Extended thinking budget for Amazon Bedrock - A new field in the advanced parameters of Amazon Bedrock AI Models sets the token budget for extended thinking on Claude models. The value is configured once and applies to Knowledge Base and AI Agent nodes alike. See Amazon Bedrock for details.
Improvements
-
One rule for optional AI Model parameters - Optional parameters (such as Temperature, Top P, penalties, reasoning effort and extended thinking budget) now follow the same rule for every provider: an empty field means the parameter is not sent and the provider default applies, any value you enter is sent as is. New AI Models start with these fields empty. See AI Models for details.
-
Provider errors in the test console - The AI Model test console now shows the errors returned by the provider without filtering, using the same parameters the flow nodes send at runtime, so you can verify a configuration before deploying. See Intents, Entities & Testing for details.
Bug fixes
- Fixed an issue where the AI section was shown in ConvyAdmin even when the Knowledge Base integration was disabled.
- Fixed an issue where file uploads and generative AI requests from the AVA interface could stall.
- Fixed an issue where conversations were not queued correctly when the agents they were reserved for disconnected.
- Fixed an issue where the image preview was not displayed in the Agent Desktop message bar.
Breaking changes
⚠️ Breaking change: New AI Models no longer come with pre-filled optional parameters (for example Temperature
0.7, Top P1, penalties0.0). If you rely on a specific value, enter it explicitly when you create the AI Model. Existing AI Models keep their saved values.
⚠️ Breaking change: On Amazon Bedrock AI Models, Completion model is now a free-text field and the model drop-down is removed. The platform no longer disables Temperature and Top P for Claude Opus 4.7 and 4.8, nor makes them mutually exclusive on Claude 4.5 and 4.6. Configurations previously prevented are now possible and are rejected by the provider at call time. Leave a parameter empty if the model you use does not accept it, and verify the configuration in the test console before deploying.
Upgrade notes
AI Models already configured keep their saved values, including values inherited from the previous defaults, and keep sending them. No migration is required. AI Models created before and after this release can therefore differ: optional fields are filled in the former and empty in the latter.