Demo Video
Watch the demo video for an overview of create tools to enhance your agent’s workflows and capabilities.
Steps to Create a Custom Tool
Access the Tools Section
From the Beam Studio dashboard, navigate to the Tools section on the left-hand side.
Here, you can browse existing tools or create a new custom tool by clicking on the Create Custom Tool button.
Initiate Tool Creation
After clicking Create Custom Tool, you’ll be prompted to define the tool objective. This is a brief description of what you want the tool to achieve.
Example objective: “This tool generates a Product Requirements Document (PRD) by taking key inputs such as business requirements and feature descriptions.”
Once you’ve entered the objective, click Generate Tool Details.
Configure the Tool
On the Configure Tool screen, you will provide essential details that define the tool:
Tool Name: Enter a name that clearly represents the tool’s function (e.g., “Product Requirements Document Generator”).
Tool Category: Select a category such as “Customer Support”, “Sales”, or in this case, “Product.”
Foundational Model: Choose the foundational model that will drive the tool. For generating a PRD, GPT-3.5 is ideal.
Tool Description: The description will be auto-filled based on the objective, but you can make adjustments to specify the tool’s functionality in more detail.
Tool Prompt & Verification
The tool prompt will be auto-generated based on the task instructions and provided inputs from the tool configuration. The prompt defines how the AI will behave and which inputs to use. While the prompt is automatically created, it’s essential to verify that it aligns with the workflow template and that any necessary changes to the workflow or tool inputs are reflected.
Verifying the Auto-Generated Prompt
Ensure Alignment: Verify that the prompt aligns with the workflow template and the tool’s purpose. If modifications have been made to the workflow (e.g., adding or removing inputs), ensure that these updates are reflected in the prompt.
Make Necessary Edits: If the workflow requires additional changes (e.g., adding new input fields or updating output formats), edit the tool prompt accordingly to ensure consistency across the workflow.
For example, if a new input such as “Legal Compliance Requirements” is added to the workflow, ensure the tool prompt includes this input and that it is reflected in both the extracted arguments and output structure.
Extracted Arguments
Extracted arguments define the key data that the tool will use to generate the output. For example, the PRD generator will extract inputs like the Business Requirement Document, Feature Description, and other key details to create a structured PRD.
Example extracted arguments:
Business Requirement Document: #businessRequirementDocument
Feature Description: #featureDescription
Target Audience: #targetAudience
Security Requirements: #securityRequirements
Compliance Mandates: #complianceMandates
Ensure that the extracted arguments match the workflow steps and inputs being passed to the tool.
Enable Memory Lookup
The Memory Lookup feature allows the tool to access information from previous steps or interactions. This is useful when the tool needs to reference earlier parts of the process, such as past PRDs or discussions about business requirements.
Foundational Model Selection
Select the appropriate Foundational Model based on the task at hand.
When selecting a foundational model, choose GPT-3.5 for general tasks, GPT-4 for complex content generation, or Claude 3 Opus for advanced comprehension. Use Mistral and Llama models for lightweight, efficient tasks, and GPT-4o for tasks involving both text and images.
Standalone Execution & Workflow Integration
Once you have configured the tool, you can test it in two ways:
Standalone Execution: Input sample data manually and run the tool to ensure it produces the expected output (e.g., generating a PRD based on mock data).
Workflow Integration: Add the tool to the workflow by assigning it to a specific step. This will allow the tool to execute automatically when the workflow runs.
Batch Execution
Batch execution is a process where a large task is broken down into smaller, more manageable parts, which are then processed in groups (batches) rather than all at once. This allows for more efficient handling of complex or large-scale operations. Instead of performing the entire task in one go, the system processes each batch separately and then combines the results once all batches are complete.
To enable Batch Execution:
Toggle the Batch Execution option during tool configuration.
Ensure the tool is optimised to handle multiple inputs simultaneously.
Finalising and Saving the Tool
After configuring the tool, click Save Configuration. The tool will then be available in the Prompts section, where you can add it to workflows or use it across agents within the workspace.
