This article will walk you through the process of creating and customizing your first AI Grid Agent.
Getting started
To begin, navigate to the AI Agent section of the platform. From there, select "Create a New Agent" to start building your grid.
From there, you will be able to set your AI agent in a Grid view, the central feature of AI Grid Agents. It allows you to visualize, manage, and analyze both internal and external data in a structured table format.
Setting up your AI Grid Agent
You can start by creating the columns
Columns Settings
Columns are the building blocks of your AI Grid. They define the structure of your data and determine what information is retrieved and analyzed for each row through prompting.
There are two types of columns:
a) Input Columns
These are text fields that you can use to include manually text data. They are generally used as data in other columns as inputs for the prompt (referenced using @ColumnTitle) and are important for variables (e.g., a Company Name that might be different for each row).
Inputs are particularly useful when working with web-based tasks.
For example, if you want to conduct a company benchmark study, you can create a "Company" input field (and add the necessary columns). When you enter your prompt (as described below), you can add the input in each prompt with "@". It will display the list of inputs you have created (in this case, the input "Company"). Simply select the input for the column. Once saved, the grid will display the Input to fulfill in the first column. You can easily add rows and enter the names of each company to perform the benchmark.
Note: Input columns are only used to include manually a text by row. You don't need to add an input to add a file as a source of data for the AI agent. When you select a tool such as Files Search or Files Analysis, a "Document" column will be added automatically.
To add inputs, simply click on the "+ Add Input" button and enter the title of the column. Column title should only contain letters, numbers, hyphens, and or underscores.
b) AI Columns
AI columns are columns where you instruct the AI agent to perform specific tasks, such as retrieving insights or searching the web. Each column you create will be a fundamental part of your grid, helping to organize your data and capture the information you need. Each column can be customized to meet your needs.
1. Column title:
Start by defining the column title, which should clearly reflect the type of data you want to collect or analyze.
2. Tools:
Tools enable you to enhance an assistant with various capabilities, allowing it to efficiently tackle complex queries.
By default, the Column is set for “Instructions Only”.
a. “Instructions Only”
- Purpose: Generate summaries, insights, or transformations based solely on LLM instructions.
- Use cases: Summarizing text, Analyzing data, Writing emails, Transforming local data, etc.
-
Best used when:
- You need to work with data already in the table, or you want to transform or analyze existing information. It can be used for example to analyze AI outputs of previous steps by using “#” (see below). Include all AI outputs of previous steps and ask for a summary, a detailed review or assessment or risks for example.
- No external information is needed. Doesn't require external files or web searches.
b. “Web Search”
- Purpose: Browse the web for up-to-date public information.
- Use cases: Gather real-time information, competitor analysis, looking up news, etc.
-
Best used when:
- You need real-time external information;
- Information might change frequently;
- You need to verify or update existing data.
Note: This tool can be combined with other tools in workflow, such as Files Analysis or Files Search. In this case, the answer of the AI agent will be based on the insights it finds in the selected documents and the web searches.
c. “Files Search”
- Purpose: Extracting detailed information from files. The tool will search for the specific information asked.
- Use cases: Extracting specific and known information from documents.
-
Best used when:
- You want to extract structured or known information from files.
Note: When selected, the AI will search inside the documents selected for the row. We're currently adding more granularity to select by columns the document to use as source of data.
d. “Files Analysis”
- Purpose: Analyze uploaded files. The tool will analyze all the content to answer.
- Use cases: Converting unstructured data into insights, summary of a file, assessment of a file, review deals, compare call insights, etc.
-
Best used when:
- You need to analyze entire document content
- The information you’re looking for might not be in the document.
Keep in mind: Each tool serves a specific purpose in the workflow. It should be chosen based on the type of data needed, whether external information is required and the task you're working on. If you want to retrieve information from a document, you always need to select either Files Search or Files Analysis tool.
3. Output format
Choose the output format for each column to ensure your data is structured correctly:
- Text: For descriptions (e.g., company activities, agreement topics).
- Word: For short description (e.g., name of founders, name of the contract)
- Boolean: For true/false results (e.g., compliance, verification)
- URL: For web addresses (e.g., company websites, LinkedIn profiles).
- Percentage: For percentage data (e.g., margins, ratios).
- Number: For numerical data (e.g., employee count, revenue).
- Date: For dates (e.g., agreement dates, founding dates).
- Email: For extracting email addresses.
- Currency: For mentioning currency.
- Single Select: Choose one option from a predefined list. When you select this format, you need to provide also the predefined list options.
-
Multi-Select: Choose multiple options from a predefined list. When you select this format, you need to provide also the predefined list options.
For the options, spaces between words are not allowed. Use “-” or “_” between words.
4. Context (optional)
You can add specific context to help the AI agent to perform the task. This can be useful if you want to explain in detail how the agent should answer, or if you want to provide context that could help to perform the task.
5. Prompts
Prompts are the tasks or instructions you give to the AI agent, such as retrieving insights or analyzing specific inputs. For more about prompting, please refer to our best practices.
When you're defining the columns, it's important to understand that columns can use data from other columns as inputs in the prompt by using @ and #:
-
To use an input of an Input Column in your prompt:Type "@" and column name (@ColumnName). Reminder: Input Columns are only for manual date entry. A column file is automatically set when you select a tool related to files search or analysis (so no need to add column for files).
- Use cases: perfect to compare different companies or markets based on the user inputs.
-
To use AI output (the results of a substack) of an AI column in your prompt: Type # and column name (#ColumnName).
- Use cases: perfect to summarize, assess or analyse previous steps in a column.
When a column uses data from another column (by referencing it with @ or #), it's a "dependency" (i.e., the column depends on data from other columns to run). When the AI agent is running, it creates a flow of data, where information moves forward through the steps. For example:
Important rules for dependencies:
- Dependencies can only flow from left to right, i.e., a column can only use data from columns to its left.
- Avoid circular dependencies (if Column B uses @ColumnA, Column A cannot use @ColumnB).
If there is no dependency, each column will be managed independently.
Tips: If you want to summarize all the previous step results, you need to include all previous AI outputs in the prompt of the column with dependencies (e.g., #Company Name, #Company Size, #Main Products, etc).
6. Select the LLM of your choice
With Lampi AI, you have access to the most capable reasoning models. For each of your column, you can select the most relevant Large Language Model (LLM).
7. Activate or deactivate LLM-as-a-Judge (beta)
LLM-as-a-Judge is an advanced evaluation feature that allows you to assess the quality of AI-generated outputs directly within your grid.
When enabled, this feature leverages aLLM to automatically review, score, and evaluate the response generated in a specific cell, based on a predefined evaluation prompt.
This is particularly useful when you're working with open-ended answers or qualitative data, and want a second layer of assurance or automated quality control.
8. Call-to-Action
You can also set specific call-to-actions for each of your columns, such as sending a mail or a message in different applications.
AI Agent Settings
- Name your AI Agent to easily find it in the list of agents and provide a description related to the goal of the agent.
- Add instructions to your agent. These instructions, or 'system prompt', guide the AI agent in completing the grid. The template includes instructions by default, but you can tailor them to suit your specific needs. For more detailed guidance on crafting custom instructions, refer to our article on creating AI assistants.
- Select the language of answers of the AI agent.
Running
Once all your Columns are set, you can start running the AI agent. For more information about how to run your AI agent, you can refer to "How to run an AI Grid Agent?".
Note that if you have selected the tool Search File or Search Analysis, you need to add a file before launching your agent. For more about how to add files to your grid, please refer "Add data to AI Grid Agents".
If you want to edit your agent, you can always return to edit or modify your grid.
Save your AI Grid Agent
Once you’ve added all the necessary columns and configured the settings, you can save your AI Grid Agent.
When you save it, you can make it public to your team and save it in a specific category.
Note: We are currently adding new ways to create tables, notably by prompting AI to create a table or by uploading CSV.