Workflows AI agents

Lampi AI offers a flexible framework for executing tasks in a structured manner. This guide outlines how to effectively implement these processes to ensure efficient task execution and project completion.

​Lampi AI Workflow is a powerful feature designed to streamline the creation and management of AI workflows, providing a robust framework for building sophisticated AI automations. It allows you to create structured workflows and implement multi-step processes that leverage the full potential of Lampi AI's capabilities.

Workflow is particularly helpful for repetitive tasks or complex tasks that can't be handled in a single request. Workflow allows you to break down complex tasks into smaller, manageable subtasks, and let AI agents perform it. You can define all the steps that our AI agents need to perform to execute your specific workflows or tasks.

Why Workflow?

• Accuracy: Each subtask gets Lampi AI's full attention, reducing errors.
• Clarity: Simpler subtasks mean clearer instructions and outputs.
• Traceability: Easily pinpoint and fix issues in your prompt chain.

When to use Workflow?

Use Workflow for multi-step tasks like research synthesis, document analysis, or iterative content creation. Creating a custom tool allows you to automate specific tasks or workflows tailored to your needs. You can define all the steps that our AI agents need to perform to execute your specific workflows or tasks.

How works Workflow?

Workflow is the perfect features to automate structured tasks. Based on advanced chain-prompting methods, it allows you to retrieve specific insights from multiple sources by section and condensate it with specific instructions. 

Workflow includes: 

• Sections: Sections allow you to structure and segment the steps of the agents.
• Prompts: Prompts are instructions to give to the agent. In the case of a Workflow, the purpose of the prompts is generally to search for specific information in the data sources.
• Tools: Tools are capabilities for agent and include "Search" and "Web search and browse" tools. For each prompt, you can select the data sources for the AI agents by clicking on the relevant icon:
  
  • Search: The Agent will perform a semantic search across the selected data to find the most relevant content. Lampi will retrieve information from the selected context in the chat where you launch the agent. Find more information about how to select context in How to manage context?;
  • Web Search: The Agent will perform a web search and browse web pages to gather information.
  • Search and Web Search: The Agent will perform both a semantic search and a web search.
  • You can choose not to select a tool. In this case, the AI agent will use the model to generate an answer to the question. 
• Condensate Prompt: The purpose of the Condensate Prompt is to analyze and consolidate all insights that the Agent has retrieved by executing all the prompts through the tools. 

Warning: The quality of the output depends largely on the quality of the Condensate Prompt. Be careful when writing it, and make your instructions as clear as possible.

Exemple of Condensate Prompt: You are an experienced market analyst, specializing in providing detailed, insightful, and data-driven information.

By using all the context, your task is to exploit, combine and include all the information and details mentioned to provide a global market overview. Your answer should present an overview of the market.

Include all information and all details mentioned in the context. If the context mentions data, KPIs, entity names, relevant examples, references, etc., it's important to keep.

Ensure your response is logically structured, and really detailed. Each paragraph should address a single main idea. Make your text easier to read by breaking it into paragraphs or line breaks. Add lines between your text. Enter spaces before the line break to begin a new paragraph. Keep a human and professional tone. Format your response using markdown. Divide your answer into labeled sections. Use clear headings (H2) and put numbers in front of headings.

You can add more relevant information and context in the prompt, for example information related to the section to draft or the purpose of the analysis.

How to create a Workflow? 

Follow these steps to create your custom tool:

1. Create a name and description

Enter a clear and concise name for your AI agent that accurately reflects its purpose. This helps in easily identifying the agent among others.

Write a detailed description of your AI agent. Explain what the agent does, the problems it solves, and the benefits it offers. Be as specific as possible to provide a clear understanding of its functionality. If you decide to create a tool with AI, Lampi will use your description to propose an action plan for the specific task.

2. Define the category and accessibility

• Category: Choose the relevant category for your agent. This helps in organizing AI agents and making them easier to find.
• Accessibility: Decide who can access this tool:
  • Private: The agent is private and only you can use it.
  • My team: The agent is shared and accessible by your team members.

3. Instruct your Agent

Include precise instructions in the Master Prompt section. The more precise you are, the closer the Agents’ behavior will be to the desired outcome.

This Master Prompt is the instructions that will be given to the agent when it will execute all the steps (except the Condensate Prompt). 

If all the steps require retrieving insights from multiple sources, instruct the agent to only answer each retrieval step by using solely the information in the context (to avoid hallucinations). 

Exemple of Master Prompt: You are an AI assistant tasked for answering queries.
Your goal is to provide accurate and relevant answers. If the query contains context, please, follow these steps:
1. Analyse the query and the context carefully.
2. Identify information and insights relevant to the query. Be sure to use and analyse all the context provided.
3. Formulate insightful, extensive, complete and structured answers.
Solely utilize the most relevant text from the contained context to answer the question. You must include all information and all details mentioned in the context. If you do not find the answer to the question in the context, do not answer and simply mention that no information is available. Use an unbiased tone and markdown format.

4. Create a Workflow

Create a Workflow with AI

You can create a Workflow with AI by clicking on "Create an agent with AI".

Lampi will propose a specific action plan with sections and related questions for the specific task based on the description you provided. 

You have the flexibility to review and edit any outputs from the AI to better fit your requirements. Customize sections and questions as needed to enhance the tool's effectiveness.

Create a tool manually

You can manually create your own sections and questions by clicking on the "+" button. This allows you to tailor the tool precisely to your specific needs.

Create a tool with CSV

You can upload a CSV with all your sections and questions. The CSV needs to be structured as follows:

• Section: indicate on each line the name of the section
• Question: indicate your prompts
• Apps: indicate the sources of context for the LLM:
  - For context search, indicate "search_context"
  - For web search, indicate "search_web".
• Condensate prompt: indicate your condensate prompt (as indicated below).

Template:

Important tip using variables: Important tips using variables

To create variables that may change each time you perform the tasks, use double accolades {{variable}}.

Example: If you want to perform a company analysis for different companies each time, write your prompt like this: Provide if available a highly detailed presentation of all the services and products offered by {{company name}}. 

Once you access the agent, you will simply need to fill in the variables (e.g., the company names).

Warning: Keep in mind that each prompt is independent, so draft them carefully to maintain context and clarity.

In the example of a company analysis, it’s important to always include the company name (or a variable) in each prompt to ensure consistency and accuracy.

5. Save the Workflow

Save the workflow by simply clicking on the "Save" button.

​Best practices for Workflows

• Order matters: Arrange sections in a logical sequence.
• Clear instructions and prompts: Provide detailed descriptions for each prompt (Master Prompt, Prompts, and Condensate Prompt) to guide the agents effectively.
• Be specific for each prompt: Keep variables or specific company name in each prompt, as each one is processed independently by the agent.
• Handle the numbers of prompts: The purpose of the condensate prompts is to analyze all insights retrieved from your prompts. If you put too many prompts in a single section, the AI agent might struggle to manage all the data retrieved. 
• Mimic your workflows: Before creating an agent, think about how you would manage the task (which information to search and where, in which order, etc) and instruct the agent to do the same through the workflow.