- Version: 2.56
- Release date: 29th October 2024
New Features and Enhancements
Function Support Within the Chat Completion Flow Action
The new Chat Completion V2 flow action replaces the previous flow action with the same name and has super-powered AI chat completion. The main enhancement is function support, which are Tenjin project flows that can be called when the large language model (LLM) detects certain user inputs, connecting to external tools or perform preset workflows. Once the project flow has completed, the resulting call and response is added to the Message Memory for use within the conversation.
There is comprehensive and detailed inline help within Tenjin that further describes the upgraded flow action, which we should strongly recommend reviewing. To help get you started, the new flow action introduces the following new configuration parameters:
- User Query: A string property that contains the user’s query (what they asked in Tenjin).
- Initial Message Memory: The Initial Message Memory is copied to Message Memory the first time the action is executed; this allows you to configure an initial system prompt or an array of role prompts. The Initial Message Memory can either be a type of string or array. When it is a string, it will be added to the message memory as `Role:System`.
- Message Memory: This is a string property that names a variable containing an array of all messages, managed internally by the flow. This replaces the need for the flow developer to maintain their own array and simplifies your flows.
There will be further updates in this area coming soon, which will upgrade the LLM chat completion flow action - watch this space!
New Flow Action
- Chat Completion v2 - This new chat completion flow action brings many advantages above-and-beyond function support; it now manages message memory for you (the user's conversation history), bringing it in line with other similar flow actions and meaning you no longer need to declare and continually update this yourself, simplifying your flows and making them easier to maintain.
Flow Action Deprecated
-
Chat Completion - The original chat completion flow action has been deprecated, replaced by the new chat completion v2 flow action as described above. We strongly recommend updating your flows to use the new flow action; the original chat completion flow action will now appear with the following message: "This action is deprecated". Whilst completing the migration, please be aware of the following:
- Previously, the Chat Completion flow action had an array property called Chat that was used to store a message array. Initial system messages and the user query all had to be added to the array before the Chat Completion action would work, and the flow developer had to manually maintain this array by adding messages returned from the flow to the array.
- Chat Completion V2 now does all this in the same way as the Query Documents Using AI flow action, it splits the Chat property into the following, as described in more detail above:
- User Query
- Initial Message Memory
- Message Memory
- The changes mentioned will require a change to your flow’s design so that it uses the updated version. When the flow action is used as a simple ‘one and done’ chat completion call, the changes will be minimal, the flow developer will not need to bundle the user query and initial messages into an array variable, instead they can simply use the User Query and Initial Message Memory fields. For more complicated flows, where the flow loops and continuously calls chat completion, you should do the same splitting for User Query and Initial Message Memory, remove the code that maintained the message array, and instead use the new Message Memory variable. If you need any assistance with the migration of the chat completion flow action, please contact us.
Upgrade Tenjin's GPT Model to GPT-4o Mini
The upgrade from GPT-3.5 Turbo to GPT-4o mini further enhances Tenjin's AI credentials and natural language understanding, improving its accuracy of answers to questions along with more intelligent conversations.
Minor Updates / Bug Fixes
- Recertify the Tenjin App in ServiceNow to support the Xanadu release of ServiceNow; the Tenjin App supports the following ServiceNow releases: Xanadu, Washington DC, Vancouver, Utah.
- Optimise flow endpoints, improving responsiveness when saving and executing flows, and reducing the risk of timeouts occurring when there are a large number of projects and flows.
- Increase the number of QnA questions displayed in read-only mode from 5 to 10, improving the ability to quickly review the knowledgebase.
- Resolve a bug with the Conversation History search, which would return no results when specific reserved characters were used; it now supports all characters, allowing you to find your users conversations.
Share this article
Comments
0 comments
Article is closed for comments.