Introduce Configurable Initial Instruction for Customizing Chat Behavior

[Mukku27]

---

Title: Introduce Configurable Initial Instruction for Customizing Chat Behavior

---

Environment:

• Feature Area: Chat Configuration for OpenAI-Compatible Models
• Version: v1.3.7

---

Issue Description: In many use cases, the ability to define an initial instruction or context for a chatbot at the start of each session is crucial for tailoring interactions to specific needs. Currently, there is no straightforward way to inject a default contextual setup or behavior guide that the model will adhere to across all sessions.

---

Proposed Feature:

• Configurable Initial Instruction: Implement a mechanism where users can define a default instruction or context that will automatically be applied at the beginning of each new conversation session. This would function as a pre-set directive that influences the model's responses throughout the interaction, allowing users to shape the chatbot’s personality, tone, or focus without manual intervention each time.

---

Benefits:

• Personalized Interactions: Users could ensure that the chatbot aligns with specific requirements or preferences right from the start, improving the user experience.
• Efficiency in Repeated Tasks: For workflows where a consistent approach is necessary, such as customer support or educational tools, this feature would save time and reduce the likelihood of user errors in setting up the chat environment.

---

Implementation Considerations:

• UI/UX: This feature could be added to the model's configuration settings, allowing users to define their initial instruction through a simple input field.
• Compatibility: Ensure that the configurable instruction does not interfere with the model’s ability to process subsequent user inputs accurately.

---

Next Steps:

• Assess the technical feasibility of adding this feature without impacting existing functionalities.
• Prototype the feature and gather feedback from users who require customized chat interactions.
• Iteratively refine the feature based on user feedback and performance metrics.

---

Attachments:

• None at the moment, but conceptual UI designs for this feature can be provided if needed.

---

[cosmic-snow]

Hi, not sure how you envision this, or why you think that is not currently implemented already. But have a look at System Prompt and maybe also Prompt Template in the models settings.

If you want several different presets per model, that's what the Clone button is for.

Also, is that request LLM generated? Sounds very artificial.