Skip to main content

Agentic Workflows

Introduction​

Agentic workflows refer to AI-driven processes where autonomous agents perform tasks with little to no human involvement. Unlike conventional automation that adheres to set rules, agentic workflows are flexible, responsive, and capable of managing intricate, multi-step operations in various fields. A system or application can independently carry out tasks on behalf of a user or another system by crafting its workflow and leveraging available tools. AI agents can include a broad spectrum of capabilities beyond merely processing natural language, such as making decisions, solving problems, interacting with external environments, and executing actions. This guide outlines how Copilot.live integrates with Boltic, allowing conversational interfaces to initiate complex backend workflows through AI Actions and tool-calling mechanisms.

Tool Calling​

Tool calling or function calling is the ability of AI to interact with external tools, APIs, or systems to enhance its functions. Instead of relying solely on pre-trained knowledge, an AI system with tool-calling capabilities can query databases, fetch real-time information, execute functions or perform complex operations beyond its native capabilities. It allows systems to complete complex tasks by dynamically accessing and acting upon external resources. Instead of just answering questions, LLMs with tool calling can automate workflows, interact with databases, perform multistep problem-solving, make real-time decisions and more.

Tool calling involves several key components that work together to facilitate AI interaction with external tools: 

  1. Recognizing the need: The AI model identifies when it lacks sufficient knowledge or requires an external function to complete a request. For example, if a user asks, "What’s the weather in Mumbai right now?” the AI recognizes that real-time weather data is needed, which cannot be derived from its static knowledge base. 
  2. Selecting the tool: The AI identifies the appropriate tool to handle the specific task, such as checking a current weather database. Each tool contains metadata and structured information, such as a unique tool name, description, parameters, and required input and output types.
  3. Constructing and sending a query: The AI formulates a structured request that the tool or API can understand, including the necessary arguments. This request is typically sent over HTTP to an external server, often requiring an API key for access. 
  4. Receiving and processing the response: The external tool returns data, which the AI must parse and structure to generate a meaningful response for the user. For instance, a weather API might respond with a JSON object containing temperature, humidity, and wind speed, which the AI then summarizes for the user.
  5. Presenting the information or taking action: The AI delivers the processed information intuitively or confirms that an action has been scheduled, such as setting a reminder. 
  6. Refining the search: If the user requests more details or modifications, the AI can repeat the process with an adjusted query, ensuring that it continues to refine its response based on user needs. 

Frameworks like LangChain are commonly used in tool calling. They provide an open-source framework for integrating external tools, APIs, and functions with LLMs. They help manage tool execution, input/output handling, and context-aware decision-making. 

Use Cases​

Tool calling enables LLMs to perform a wide range of tasks across various domains: 

  • AI automates workflows such as scheduling meetings, sending emails, or managing to-do lists through integrations with platforms such as Google Calendar.
  • Smart Devices and IoT Monitoring: Agentic AI systems can monitor and control automation systems, IoT devices, and robotics. One day, all end-to-end workflows will be handled by autonomous agents. 

Applications​

  • Customer Support: AI agents handle inquiries, troubleshoot issues, and provide solutions without human intervention.
  • Supply Chain Management: Agents monitor inventory levels, predict demand, and optimize logistics in real-time.
  • Software Development: AI agents generate code, perform testing, and manage deployment processes.

Benefits​

  • Increased Efficiency: Automating complex tasks reduces time and resource consumption.
  • Scalability: Agentic workflows can handle growing workloads without proportional increases in human resources.
  • Improved Accuracy: Continuous learning enables agents to reduce errors over time.
  • Enhanced Decision-Making: Agents can quickly process vast amounts of data to make informed decisions.

How it Works​

At a high level, Copilot.live is the user-facing layer, while Boltic is the backend engine.

When a user initiates a request, such as “Schedule a meeting with the sales team” or “Summarize today’s leads,” the input first reaches the Copilot microservice, which relays it to the LLM. The LLM then interprets the request, maintains context, and determines whether it can handle the task internally or needs to trigger an external workflow.

At this decision point, Copilot leverages a tool-calling mechanism, identifying and routing the task to an appropriate external service. In this architecture, that external service is often a Boltic workflow. When Copilot routes a request to Boltic, it typically sends a payload to trigger a specific workflow.

Boltic then parses the input and initiates activities that may include database queries, CRM updates, etc. Once the Boltic workflow completes its task, it responds to the Copilot, which then passes the output to the LLM for contextual interpretation. Finally, the processed response is rendered back to the user.

Copilot Tool​

The Copilot Tool allows a Copilot to communicate with an external Boltic workflow.

When building an AI Action that invokes a Boltic workflow, you’ll need to define a Copilot Tool, which includes essential metadata such as the tool name, description, and the expected input/output schema. This configuration enables Copilot to understand what kind of data to pass to the workflow and how to handle the response.

When creating a Boltic workflow to be triggered by Copilot, only two elements remain consistent across use cases:

  1. Copilot: The conversational agent that captures user intent.
  2. Copilot Tool Definition: The configuration that connects the AI Action in Copilot to the corresponding workflow endpoint.

Everything else—including triggers, activities, data transformations, and response formatting—depends on the workflow's specific use case.
Refer to Copilot Tool document for more details.

Copilot.live Documentation​

  1. Getting Started with Copilot: Learn how to create and configure your first Copilot, including setting up personality, response styles, and user flows.
    Refer to Copilot Guides to learn more about Copilot.live.

  2. Deploying Your Copilot: Explore deployment options across the web, mobile (iOS/Android), WhatsApp, and voice. Includes SDK setup and embedding options.
    Refer to Copilot Deployments document to learn how to deploy your Copilot on platforms.

  3. Build Workflows: Automate tasks like handling customer queries or managing complex processes easily.
    Refer to Workflows to learn to build workflow automations.

Boltic Workflow Documentation​

Boltic is a workflow automation engine that allows you to define modular, trigger-based workflows connected to external services. It powers backend execution for tools like Copilot.

  1. Workflow Overview: Understand the structure and purpose of workflows in Boltic, including how they’re used to automate logic and interact with APIs.
    Refer to Workflow Overview for more details.

  2. Setting Up Triggers: Learn how to initiate workflows using events such as webhooks, schedules, or manual calls.
    Refer to Triggers for more details.

  3. Using Activities in Workflows: Activities define the logic blocks in your workflow—like condition checks, API requests, or data transformations.
    Refer to Activities for more details.

  4. Workflow Version Management: Manage and iterate on workflow versions for safer deployments, testing, and rollback strategies.
    Refer to Workflow Versions for more details.


Was this section helpful?