Agents in Opentroy
Package your LLM into an Agent, providing it with tools, context, and a system prompt to perform tasks
Guide to Creating and Managing Custom Agents in Opentroy
This section provides a step by step guide on how to create and manage your own Custom AI Agents in Opentroy. With Opentroy's intuitive interface and AI-powered tools, you can build personalized agents without requiring any coding knowledge.
1. Overview of Custom Agents
A Custom Agent in Opentroy packages:
A local or distributed Large Language Model (LLM)
A System Prompt that guides or constrains the model’s behavior
One or more Documents/Files providing additional context
Tools that enable the Agent to interact with external systems, APIs, or services
By configuring these elements, you create a specialized Agent suited to a specific use case or workflow. This differs from using a default or global model because it bundles everything the model needs prompting rules, relevant documents, and tool access into a single, easy-to-manage configuration.
How to Navigate the Agent Creation UI On the main menu at the bottom left of the app, click on the AIs button with the robot icon. Then click on the Agents tab at the top. To create a new agent, click on the + Add Agent button.
2. Agent Name and Description
1. Agent Name
Purpose: Serves as the main identifier for your custom Agent.
Format: Must use lowercase alphanumeric characters and underscores (e.g., my_cooking_agent).
Why It Matters: Having a standardized, clear name makes it easier to manage multiple Agents in Opentroy, especially when they serve different purposes.
2. Description
Purpose: Offers a brief summary or explanation of the Agent’s role.
Usage: Helps other users (or yourself at a later time) quickly understand what the Agent does and why it was created.
Effect on Behavior: While the description does not directly change how the LLM responds, it is useful for guiding users to select the correct Agent.
3. System Prompt
The System Prompt provides the primary instructions or “rules” under which the Agent operates. This prompt can define:
The Agent’s “personality,” style, or tone
Specific instructions for handling conversation flows
Which tools to use and in what circumstances
References to relevant documents, data, or context
All subsequent user prompts in the chat are interpreted through the lens of this System Prompt, making it critical to craft it carefully.
4. Model Parameters: Temperature, Top-P, Top-K
When fine-tuning how an Agent generates responses in Opentroy, the platform exposes several sampling parameters:
Temperature
Controls the amount of randomness in token selection.
A higher value (e.g., 1.0) produces more creative or varied responses, while a lower value (e.g., 0.2) results in more deterministic outputs.
Top-P (Nucleus Sampling)
Sets a probability threshold to limit token selection to a “nucleus” of the most likely tokens.
Lower values make the model pick from a narrower range of high probability words, producing more focused text.
Top-K
Restricts the model to choosing its next token from only the top K likely candidates.
A lower K focuses on the most probable tokens, potentially reducing diversity.
These parameters can be tweaked individually or together to balance creativity, reliability, and relevance based on your use case.
5. Tools and Their Usage
1. Enabling Tools Each Agent can have specific tools enabled or disabled, such as DuckDuckGo Search, Google Search, or various API callers. When toggled on, the Agent can invoke these tools if it determines they are relevant to the user’s query. When toggled off, the Agent will not have access to these capabilities even if the user asks for them.
2. Creating New Tools The "Create New Tool" button in Opentroy opens a “tool playground,” where you can define custom prompts or scripts in natural language (AI-powered, requiring no coding knowledge). This reflects Opentroy's promise to enable powerful tool creation without programming expertise. Practical Use Cases: Creating specialized data-fetching routines, connecting to proprietary APIs, or automating tasks within an organization. For more detailed coverage, refer to our dedicated Tool Creation and Configuration documentation.
6. Agent Context: Adding Files and Folders
A. Purpose of Agent Context This section lets you attach local or cloud-stored documents and folders to your Agent. The system extracts embeddings from these files so the LLM can reference them for context, data lookup, or analysis.
B. How to Add Context
Upload your files to Opentroy in the AI Files Explorer section.
Select which files or folders the Agent should have access to in its configuration screen (e.g., “Set Chat Context”).
Confirm your selection so the Agent can retrieve relevant information from them during the chat session.
C. Example: “Soup Recipes.pdf” If your Agent’s context includes a PDF with soup recipes, the LLM can pull relevant recipes or cooking instructions from it when a user asks.
7. Using Agents
To use an agent, navigate to a new chat and from the same selector where you choose the model, you can now select the agent you want to use. Custom agents will appear at the top of the list.
8. Conclusion
By configuring each setting from Agent Name and Description to the System Prompt, model parameters, Tools, and Agent Context you tailor an Agent’s abilities to meet very specific requirements. This flexibility allows you to create a range of custom Agents, each optimized for different tasks or styles of interaction.
Last updated