Caddi is an AI-driven automation platform that discovers repetitive back-office tasks and allows users to build agents through screenshare and chat.
What the product does and how it is positioned
Caddi is an agentic automation tool designed to streamline back-office operations by identifying and executing repetitive workflows. It functions by analyzing the systems a company already uses to find high-volume tasks that can be transitioned to AI agents.
The platform utilizes a teaching-based approach where users record their manual processes and narrate the steps. This information is converted into deterministic code, so that the automation follows specific rules while providing a full audit trail for every action taken.
Source-supported ways to use the product
The agent reads incoming messages in Outlook, identifies document types in Docusign, and files them to the appropriate matter in NetDocuments.
Caddi matches trust payments and remittances to payables in QuickBooks, automatically posting entries or raising exceptions for review.
The documented workflow, where available
The system reads connected business applications to identify and rank repetitive tasks based on frequency and step count.
A user performs the task while screensharing and narrating the process to capture rules and logic for the agent.
The platform generates a loop diagram and automation plan, which the user can refine through a chat interface to handle edge cases.
The agent executes the task across integrated tools, logging every decision and step in an audit history for leadership review.
Caddi differentiates itself by using a 'teach like a new hire' model for automation. Instead of traditional coding or complex flowcharts, users interact with an agent named Loopy to define workflows. The process begins with the Caddi Recorder, which captures the visual and verbal nuances of a manual task.
Once a task is captured, the platform converts the recording into a structured automation plan. This plan uses deterministic code for execution, which means the AI does not improvise on the path taken; it follows the established rules exactly as taught, providing stability for regulated industries.
Checks to run with your own material and workflow
What was checked and when
Answers based on the source-checked product record
The discovery feature reads your existing systems to identify and rank tasks that repeat frequently, counting the steps involved in each.
Users train the agent by recording their screen while performing the job and narrating the steps, similar to training a new employee.
Yes, users can specify rules during the recording phase or use the chat interface to tell the agent how to handle specific exceptions.
Every run is logged in the run history and audit log, showing what the agent decided, why, and which permissions were used.
The platform is designed for regulated sectors such as law, financial services, accounting, private equity, and banking.
