⚙️Advanced Bot Settings
Welcome to the advanced bot settings guide. This guide will assist builders with customizing the bot for specific use cases and workflows. Learn about advanced bot capabilities, models, and agents.
1. Setup
Assign Categories
Create tags/categories based on the bot's expertise to help users find them more easily with filters in their dashboard.
Welcome Message
Set the default greeting message users will see when they start a new conversation with the bot. Manage expectations by telling users about the purpose of the bot.
Conversation Starters
Turn on conversation starters by enabling the toggle and configure up to four intro messages users can send to the bot. This will help users get a sense of how to interact with the bot and start the conversation swiftly.
This is an example of what the user will see when they start a conversation with the bot:
Description
Summarize the bot's purpose. That will help users understand which use cases the bot is suitable for.
Users can see the bot description when hovering over the info icon of a bot in the dashboard selection.
2. Models and Capabilities
In this section, you can select and configure the AI model that will power your bot. You can also enable or disable advanced skills like text-to-speech, speech-to-text, and image retrieval from documents.
Select LLM Model
Choose the appropriate Large Language Model (LLM) model that fits your use case and complexity requirements. The model will determine the general capabilities and response quality of your bot.
Available Models for Knowledge Bots
Last updated: February 2024
Model | Description | Costs | Max Tokens | Training Data |
---|---|---|---|---|
Azure GPT-3.5 Turbo | Great for most every tasks, faster responses, and cost efficient. | Lower | 16,385 | Up to Sep 2021 |
Azure GPT-4 | Able to do more complex tasks, but slower at giving answers. | Moderate | 32,768 | Up to Sep 2021 |
Azure GPT-4 Turbo | The most capable model, best for tasks that require enhanced reasoning and creativity. | Higher | 128,000 | Up to Dec 2023 |
Data Privacy Note
We currently provide access to LLM models exclusively through Microsoft Azure's European cloud servers for our clients, ensuring full compliance with the highest EU data security standards and GDPR data privacy regulations.
When using the GPT models from Microsoft Azure on our platform:
Your inputs, outputs, and all data you provide or generate are strictly confidential.
Your data is NOT used to improve any Microsoft or 3rd party products or services.
OpenAI and Microsoft do NOT use your data to train their models.
Unlike ChatGPT, the Knowledge Bots do not take user data to retrain any AI models or for any other purpose other than to fulfill your request. Learn more here.
Model Modifiers
After selecting your preferred LLM model, delve into the 'Model Modifiers' to precisely calibrate your bot's output. Here, you can adjust settings related to creativity, vocabulary range, topic and word variety, and search scope. These tools are designed to fine-tune the bot's responses, making them more relevant and effective for your specific use cases, from enhancing creative brainstorming sessions to ensuring accurate, focused answers in information retrieval.
Creative Freedom (Temperature)
Adjust the creativity and predictability of the bot's responses. Higher values encourages more inventive and varied output, while a lower value yields more reliable and consistent text.
Use Cases: Opt for lower temperatures when accuracy on specific topics is crucial, and higher temperatures for creative or brainstorming sessions.
Scale: 0 (more factual) to 2 (more creative).
Note: Values above 1 are not recommended as it may lead to less reliable outputs, since the AI could produce more random responses.
Vocabulary Range (Top_p)
Adjust how varied the bot's language is. Higher values leads the bot to use a wider range of words, which can make responses more varied but possibly less on-point. Lower values keep the bot's language closer to the most likely and relevant words, ensuring responses are more focused and directly related to your query.
For example, choosing a value of 0.1 limits the bot to using words that fall within the top 10% most probable, ensuring the language stays more precise and relevant.
Use Cases: Set lower for more focused and consistent answers, higher for a wider exploration of ideas.
Scale: 0 (more relevant) to 1 (more diverse).
Topic Variety (Presence Penalty)
Determine how often the bot introduces new topics. A higher penalty discourages the bot from repeating topics, promoting variety, whereas a lower penalty will result in more focus on previously mentioned content.
Use Cases: Increase for generating new ideas or themes, decrease for detailed exploration of a topic.
Scale: 0 (more coherent) to 2 (more variety).
Word Variety (Frequency Penalty)
Influence the repetition of words in the bot's output. Higher values reduce the recurrence of words, fostering a richer language, while lower values allow the bot to repeat terms more frequently, which can help maintain focus on specific concepts.
Use Cases: Use higher values to keep the language fresh and varied, lower values to emphasize specific terms or concepts.
Scale: 0 (more consistent) to 2 (more variety).
Search Range (Top_k)
Define the scope of information the system reviews from the uploaded documents to generate responses. A lower value narrows the vector search to a smaller set of highly relevant information, ensuring more precise answers. A higher value widens the search to include a more extensive selection from the documents, offering broader but also less accurate insights.
Use Cases: Lower values (0-5) are ideal for specific questions that require precise and accurate answers. Higher values (5-10) are more suitable for complex and broader questions that need to explore multiple perspectives.
Scale: 0 (focused, precise) to higher numbers (broad, varied)
Audio Output (Text-to-Speech)
Select a voice for the bot to communicate the messages verbally by activating text-to-speech capabilities, providing a more natural and accessible user experience.
Audio Input (Speech-to-Text)
Equip your bot with the ability to understand spoken language by incorporating speech-to-text functionality, allowing for hands-free interactions.
Users will be able to record a message via the microphone button in the chat interface.
Image Search and Retrieval
Enable the bot to pinpoint and retrieve relevant images from the uploaded documents, enhancing its ability to provide visual information to deliver more comprehensive answers.
3. Agents
Agents are specialized AI-driven tools designed to perform specific tasks. These tasks can range from summarizing documents to generating reports, writing emails, and more. By leveraging agents, users can automate and streamline various processes, enhancing productivity and efficiency.
Key Features of Agents:
Task-Specific Functionality: Agents are tailored to perform particular functions, such as summarizing text, generating insights, or drafting emails.
Integration with Nexus: Agents work seamlessly with Nexus, our advanced knowledge bot, to extract, create, and share new knowledge effortlessly.
Using Agents Properly Users have the flexibility to either select from a range of pre-configured agent templates or create their own custom agents to meet specific requirements. The pre-configured templates are designed to streamline common tasks such as summarizing documents, generating reports, or drafting emails, ensuring quick and efficient execution. For more tailored needs, users can customize the template by defining unique skills and abilities, thereby creating agents that precisely align with their specific objectives.
Setup your Agents
Step 1: To setup agents, navigate to the agents tab in your bot settings.
Step 2: To modify existing agent templates click the "+" to the left of the respective category you want to modify.
Step 3: To create your own category click "+ Add Agent Category" and then modify the newly created agent template.
Step 4: You can define a category & tasks names. But most importantly, meticulously describe the task in the field provided.
Default Categories
Chat Summaries
This agent lets users create brief summaries of their conversations, drawing out the key points for future reference.
Default Prompt:
Summarize the following conversation in 150 words, focusing on the key points, takeaways, and main topics discussed. Ensure the summary is concise and provides a clear overview of the conversation's content and context, suitable for reference in future interactions. Remember, the summary should not exceed 150 words and should capture the essence of the discussion, aiding in maintaining conversational continuity.
How to Use: In the chat interface, navigate to "Chat Summaries" and then click on "Generate Summary" to get a condensed recap of the conversation so far. This is ideal for keeping track of decisions, action items, or just refreshing your memory on past interactions.
Follow-up Questions
This agent lets the bot automatically suggest follow-up questions for the user after every bot response to help users make natural progress based on the chat history and context.
Default Prompt:
Generate four insightful follow-up questions that the user can ask you based on the chat history and user's idea to further refine, challenge, and understand the concept, so that the user can identify actionable next steps in the ideation process together with you.
Last updated