Laserfiche AI Agents

Laserfiche AI agents allow users to easily to automate time consuming tasks such as organizing and categorizing documents and folders. A user simply needs to tell the agent what they want it to do using natural language, and it will determine the necessary steps to accomplish it and present them to the user for approval. When the user approves the steps, the agent will automatically do it for them. The user saves time they otherwise would have needed to spend performing the task manually, without needing to take up a process designer's time designing a process. For example, the user could use a directed agent to quickly perform bulk actions that are not predicable enough or frequent enough to merit a process at all, without having to do those bulk actions by hand.

Since the agent is acting on behalf of the user, it is constrained by the same security settings as the user; see Security and Auditing Considerations, below, for more information.

See a video on directed agents

Learn how to use Laserfiche directed agents to automate bulk document actions like organizing files, renaming documents, and moving content using natural language prompts while maintaining user approval and permissions. By reducing manual administrative work and simplifying complex document tasks, AI Agents help teams improve efficiency, consistency, and operational speed without building workflows.

Using Directed Agents

Users provide prompts to directed agents using Smart Chat in the Laserfiche Repository Web Client. Select the Laserfiche AI icon () to open Smart Chat if it is not already open. Smart Chat will display its current mode (ask or agent) in the lower left corner of the chat window; if it is not already in agent mode, click to toggle it to Agent.

Specifying Entries to Act On

There are two ways you can specify which documents you want to ask questions about or perform actions on. First, you can select any folder or folders, set of search results, collection or collections, or specific individual documents. Second, you can use Smart Chat in ask mode to locate documents that meet specific criteria, and then switch to agent mode in the same chat window to perform actions on those documents. This allows you to either directly specify the documents or groups of documents you want, or to use Laserfiche AI to dynamically locate them. You can also specify you want only certain documents to be affected in the prompt itself. For example, you might select a folder, and then prompt the agent to move only the documents in that folder that have been assigned a particular template. Note that the agent can only refer back to prior information in the chat history if the chat history has not been cleared, either manually or automatically. Once the chat history is cleared, the information can no longer be referenced by Laserfiche AI.

Creating Prompts and Confirming Actions

Once you have collected your documents, you can type your prompt in Smart Chat, specifying what you would like to do with the selected entries. Agents can perform a number of tasks; see Available Actions, below, for more information. You can specify the prompt using natural language in a conversational style. If the agent doesn't have enough information to process your response, it will prompt you to clarify.

The agent will display the actions it will perform to accomplish your goal under Planned Operations, allowing you to Confirm or Cancel. If your prompt is complex or multi-step, you will be given three options: Confirm entire plan, which will take all the steps at once; Confirm next step, which will prompt you to confirm each individual step; and Cancel. (Note that for some operations, you will always be required to confirm each step.) If you choose to confirm each step individually and then cancel after taking one or more steps, the steps you confirmed will be applied as you confirm them, but no additional steps will be taken. If you chose to confirm the entire plan, you can select Pause to stop the agent at the current step, and then confirm further actions individually.

Once you have confirmed, the agent will take the appropriate action or actions, and then will provide you with a summary of what was done.

The specific actions that a given user can instruct an agent to take are limited by their security settings. See Security and Auditing Considerations, below, for more information.

See a video on best practices for Laserfiche directed agents

Learn best practices for writing clear, structured prompts that help Laserfiche directed agents perform repository tasks accurately and predictably.

Available Actions

AI agents can perform a variety of tasks in your repository. See the links below for instructions and examples for each action directed agents can perform:

Examples

The following examples demonstrate ways you can use directed agents to perform both simple and complex operations. You can also see additional examples for each of the available actions by following the links above.

See a video highlighting example scenarios for using directed agents

This video demonstrates how Laserfiche directed agents use natural language prompts to automate complex repository tasks such as content cleanup, metadata validation, public records request preparation, contract management, and payment risk analysis. By automating repetitive work while keeping you in control of every action, directed agents help improve accuracy, reduce risk, and free your team to focus on higher-value work.

Single-Operation Examples

Sorting Documents by Name

Joe wants to sort all documents with the word "Expedite" in the name into the folder "Expedited Cases." He selects the folder he wants to organize and uses the prompt "Move all documents in this folder that have "Expedite" in the name to the subfolder "Expedited Cases"." The agent analyzes the documents and provides a list of which documents to move for him to confirm. Upon confirmation, it moves the documents, creating the new subfolder if necessary. This saves Joe time locating and then manually moving the documents to the folder.

Standardizing Document Names

Maria has imported a large quantity of PDFs that have nonstandard names. She wants to clean them up by changing dashes and underscores to spaces, and by changing them to title case. She selects the documents she wants to rename, opens Smart Chat in agent mode, and uses the prompt: "Rename the selected documents to meet the following criteria: all underscore (_) and dash (-) characters replaced with spaces, and the names in title case." The agent analyzes the selected documents and gives her a list of new names to review. When she confirms, the agent renames the documents.

Populating Field Values

Some fields applied to agendas have not been filled in, and Pat wants to quickly correct that so that they will be more readily searchable. After selecting the documents, Pat uses the prompt "Fill in the "Date" field with the date of the meeting the agenda pertains to, and the "Summary" field with a summary of no more than 200 words listing the topics on the agenda." The agent presents a list of documents for which these field values will be extracted. When Pat confirms the actions, the agent populates those fields based on the contents of the agenda documents.

Sorting Documents into Folders by Date

A user, Joaquin, wants to organize documents in their personal folder into subfolders by the year they were created. These folders do not yet exist in his folder. Joaquin selects the documents in the folders, opens Smart Chat in agent mode, and uses the prompt "Move documents to folders based on the year each document was created." The agent presents him with a list of folders that will be created and the documents that will be moved into those folders. When he confirms the actions, the agent creates a folder for each relevant year and moves the correct documents into them.

Renaming Documents Based on a Field

Linda has a large number of documents with inconsistent or difficult-to-understand names. She has already used Smart Fields to extract more appropriate names from the document contents and stores the names in a "Title" field. She now wants to rename the documents using the same name as that field's value. To do so, she selects the documents she wants to rename, opens Smart Chat in agent mode, and uses the prompt "Rename selected documents using the value in the "Title" field." The agent presents her with a list of new document names to rename them to. When she confirms the action, the agent renames the documents.

Multi-Operation Examples

Contract Management

Joel manages contracts, and wants to find a way to identify which contracts need reviewing. To that end, he wants to identify active contracts that do not have an owner listed as well as contracts that expire within thirty days. These contracts all have the "Vendor Contract" template applied, and have fields with information about the document's status (active or inactive), the listed owner, and the expiration date. His repository already has the tags "missing owner" and "expires within 30 days." Joel wants to apply the "missing owner" tag if the relevant field is blank, and the "expires within 30 days" tag if the expiration date is within thirty days. He then wants to create a folder and place shortcuts to these tagged documents in it for easy review by his team.

Joel selects the folder "Contracts" and opens Smart Chat. Since this folder has subfolders and he wants to perform this action on subfolders as well, he ensures that the Include the contents of the selected folders button is selected. He switches Smart Chat to agent mode. He then writes the prompt: "For the active contracts only, if the owner is blank, apply the tag "missing owner" and if the expiration date is within 30 days, apply the tag "expires within 30 days". Create shortcuts for any files you tagged in a folder called "For Review". The agent will first analyze the documents, identify those with a blank "Owner" field and/or an "Expiration Date" field with a date within thirty days, and apply the specified tags. The agent provides him with a list of documents that meet those criteria and the operations it will perform on them. Since this operation has multiple steps, he will be prompted to choose whether to confirm the entire plan, or to confirm each step individually. Joel chooses to confirm the entire plan. When he confirms, the agent will tag the relevant documents, create a folder called "For Review," and create shortcuts to the tagged documents in that folder.

Note that Joel was able to use natural language in his prompt. He did not have to specify that active contracts are defined by a specific field value, or that owner information is stored in the Owner field, as the agent can locate that information itself. The agent was also able to perform the calculation to determine which documents were expiring within 30 days, as the agent could determine that itself based on the expiration date field. He also did not have to pre-create the "For Review" folder. These operations were performed based on his natural language prompt, allowing him to quickly perform this complex operation over many documents, without doing it all by hand.

Flagging Potential Risks

Sofia needs to monitor contracts for Internal Control over Financial Reporting (ICFR) risks of the type mandated under SOX 404. An "ICFR Risks" tag already exists in the repository. While Sofia knows that she will need to review the documents herself to determine the actual risk, she can use an agent to identify and categorize these documents for a first, exploratory pass.

Sofia selects the folder "Contracts" and opens Smart Chat. Since this folder has subfolders and she wants to perform this action on subfolders as well, she ensures that the Include the contents of the selected folders button is selected. She switches Smart Chat to agent mode. She then writes the prompt: "Identify potential ICFR risks among active contracts, of the type mandated under SOX 404. Tag these documents "ICFR risk" and add to a new collection." The agent will display the new collection that it will create to contain these documents. Since Sofia did not specify a name for this collection, the agent will prompt her to do so. When Sofia has confirmed the collection name, it will prompt her to choose whether to confirm the entire plan or confirm the next step only. Since Sofia wants to review these actions before they are completed, she chooses to confirm only the next step. She will then be prompted to confirm the set of documents to be tagged, the creation of the new collection, and adding the tagged documents to that collection.

In this case, note that Sofia did not need to explain what ICFR or SOX 404 are; as these are standard industry terms, the agent understands them in context. If Sofia was referring to a non-standard term (for instance, something specific to her organization), she would need to explain it to the agent. She also did not need to provide names for the new folders, as the agent could determine relevant names from context (and if Sofia did not like the names it came up with, she could provide new names at the confirmation stage).

Information Available to Directed Agents

Directed agents have access to the following information about your documents to analyze documents and perform actions:

Entry name
Template
Field values
Tags
Document relationships
Entry type (document, folder, or shortcut)
Entry path
Page count
Creation date
Created by
Last modified date
Last modified by
Whether the document is under version control

Security and Auditing Considerations

For security and auditing purposes, direct agents act as extensions of the user that is using them. The agent can only take actions that the user themselves could. For instance, if a user has sufficient rights to move a specific document to a folder, then the agent can do so on their behalf. If the user lacks any of the necessary rights to move the document, however, the agent will not be able to perform the action either. In effect, the agent is performing the act on behalf of the user, and therefore has only the access the user has.

Similarly, when an action is performed by a directed agent, the operation will be audited as performed by the user who prompted the directed agent. Again, as the agent is acting as an extension of the user, the audit logs will reflect that fact. Auditing also distinguishes actions the user approved individually from actions auto-approved after the user confirmed the entire plan.

For more information on Laserfiche AI security and governance, see the Laserfiche Security, Trust, and Compliance page.