Skip to content
Wingbot Chatbot Framework Documentation

wingbot.ai LLM features

Supported from v3.70.0

Terminology

  • Instructions - High-level configuration or rules that define the behavior, tone, or goals of the AI agent. Typically static and authored by developers (e.g., "You are a helpful assistant").
  • User Input - The most recent message from the user — a question or message that the LLM is expected to respond to.
  • Chat History - A rolling log of previous user inputs and AI responses, used to maintain continuity in multi-turn interactions.
  • Response - The generated reply from the LLM based on the Instructions, Chat History, and current User Input or Command.
  • Prompt - The full constructed request sent to the LLM — typically includes Instructions, Chat History, and the current User Input or Command.
  • Command - A description of a task to be performed by LLM. Mostly used for internal prompting such as result validation, translations, e.t.c.

Configuring a testing chat to use ChatGPT

Testing chat environment now offers ability to insert LLM token for accessing ChatGPT APIs. Then it’s possible to develop an AI agent on a simple testing environment and to leverage all advantages of Wingbot’s Hybrid AI.

image.png

Sending a simple Prompt

Converting a Message into a Prompt

To simplify conversion of predefined interactions to LLM we introduced a switch to convert a text message into a prompt with all current features of a message.

Use the caret by a message to expand options and switch it to prompt as instructions.

image.png

LLM Instructions

Don’t repeat yourself

To share LLM instructions across all the prompts within a bot simply put the “LLM Instructions” resolver into a Middleware interaction:

  • at the top of a dialogue
  • mark the interaction with a star •

image.png

Instruction set chain

As the bot executes user Input, it puts all the visited instructions into a single system instruction (joined with a double newline)

  • globally: at the top of the Root dialogue
  • locally (in a dialogue): at the top of the specific dialogue to add additional details for a given dialogue to your instructions

When the instructions of a specific set are put consecutively, they’re combined in a given order one after another. Instructions in the Root block first, then the instructions at the top of a dialogue.

Including one instruction set in another

Instructions of a one set can be included i another set by following syntax: ${<instructions set name>}. For example, when you want to use a default instruction set in a different set, just use ${default}.

Using the last instruction as a part of instruction set

For example, if you need to put the last instruction into brackets, you can use the ${last()} syntax. The last instruction would’t be appended to the last instruction, but will be placed instead of the syntax.