This is done through a text-to-workflow capacity that allows even nontechnical users to describe a workflow in natural language and have Omni build and run it. For instance, a user could type "when a new client is added in Karbon, pull their entity details, connect their QBO file, ingest the last 12 months of transactions, run a diagnostic to surface reconciliation gaps and send the assigned manager a health report on Outlook before their onboarding call" and Omni will generate an auditable, GAAP-compliant agentic workflow automatically, including any integrations to relevant tools.
There is no need for coding, or implementation assistance. Qumarss Bagheri, chief of staff for Artifact, said in an email that Omni's Integration Agent can build integrations on the fly to any tool which has API documentation (whether that be internal or publicly available documentation). He added that this does not just connect tech stacks but enables workflows to run across and through tech stacks, orchestrating systems in a way that does not so much replace these solutions but, rather, uses them all (e.g., AP/AR tools, ERPs, communication tools, practice management systems, internal, industry specific tools, etc.) to develop workflows.
Workflows are modular, he added. Users can, in minutes, reconfigure them for specific clients, specific industries, specific departments or a whole firm's use (e.g., a workpaper generator template for real estate practices could output a report of type X into Outlook instead of a report of type Y into Slack). Meanwhile, every Omni workflow feeds corrections back into Arti's models at the client, industry and firm level, so each close cycle requires less manual intervention than the last.
Omni is not intended to be a replacement for other tools, but rather something that orchestrates and coordinates their use across the workflows. On a practical level, this might mean a month-end close workflow might need to pull transaction data from QuickBooks Online, match it against bank statements, cross-reference invoices in Bill.com, check payroll entries from Gusto, generate a workpaper in Excel, and flag exceptions to the accountant via Slack—all for a single client, and repeat every month.
"Today, a human does each of those steps manually, logging in and out of systems, transferring data across them, leveraging native integrations where possible or perhaps engaging their tech team to build bespoke integrations. Omni chains those steps together into an automated workflow that runs end-to-end, either on a schedule or on a manual trigger," said Bagheri in an email.
When a user describes what they need in natural language, under the hood the AI parses the description into a structured multistep workflow plan, mapping each step to a specific operation (e.g., connect to a data source, then run a reconciliation, then apply a threshold, then generate an output, then route an exception.) Each step is best thought of as a node in the workflow. The system then uses a multi-agent architecture to build the respective nodes, configure them accordingly and connect them according to the plan it has drawn. The system might re-use a node it already has in the system; or if there is no node for the step needed, it will build a new node from scratch. The system will present the complete workflow to the user, who can then prompt further to change the order of nodes, how nodes are configured or which nodes are used. The user can then run the workflow as a test, review it, edit it, confirm permissions and run it on demand or on a schedule. When the workflow executes, agents carry out each step in sequence, connecting to the relevant systems via API, performing the accounting work, and producing the outputs.
The name nods toward a kind of universality, as Bagheri said users can build pretty much any workflow they want with this solution, limited only by the description a user gives of it.
"With Omni, a firm can build whatever kind of workflow they can think of. The breadth of possibilities is (practically) infinite because workflows are just different combinations of 'nodes' any node can be built and nodes can be organized in (practically) infinite ways," he said.
Omni operates via API calls, essentially machine-to-machine communication, as opposed to using robotic process automation or screen-scraping techniques that simulate human keyboard and mouse actions. He said this was a deliberate choice. The developers felt API-based integrations were more reliable, faster, more secure and auditable.
Bagheri added that the system is designed for autonomous execution with human oversight versus human operation. When bluntly asked by Accounting Today whether someone could just set up Omni and then go play video games the rest of the day, he said they could.
"This is one of Omni's most important differentiators. The system is designed for autonomous execution with human oversight — not human operation. So, yes, a human can type out a big prompt, go and play a video game whilst Omni builds a workflow, come back, check things are in order, configure AOuth permissions, make changes to the workflow structure through natural language prompting, and continue to operate like this until they are satisfied with the workflow design and output. From there, the workflows can run completely autonomously, perhaps on a schedule; or they can run from manual triggering," he said.
This means that, in theory, Omni can run on a schedule in the background without human intervention. Workflows can be scheduled to run on recurring triggers (e.g., first business day of each month, every Monday morning, whenever a new bank statement is available). The system doesn't require someone to log in and press a button each day. And every run of the workflow produces a full audit for review. Further, the self-learning component means each cycle teaches the system more about a specific client's patterns, reducing the exceptions that require human attention over time.
However, it does not necessarily have to work this way. Users can decide to add human-in-the-loop or review steps in the workflows if they want to always check the result before sending it to a client, for example. Bagheri added that, in practice, humans might want to be more directly involved, at least in the beginning. Building workflows for complex tasks takes refinement and iteration to tackle edge cases at scale; while a "set it and forget it" one-shot promoting workflow is theoretically possible, he said it would be unwise and unreliable to do so.
"Ultimately, it would be a user's decision how they wanted to run their workflow. If they really wanted to, they could remove themselves from the loop entirely. This is entirely advisable for automated, repetitive tasks that don't require professional judgement," he said.
Omni will be based on a credit/consumption-based model. For large enterprise and bespoke co-builds, there will be custom pricing.
