On August 20, 2026, Serval made its intelligent agent Catalyst available to all organizations using its platform, with the agent enabled by default. The company positions Catalyst as a unified administrative layer above its IT service management platform, allowing it to analyze repetitive work, suggest what should be automated, and then prepare the components needed for the automation instead of waiting for an administrator to specify every step manually.
Catalyst can inspect support-ticket histories, standard operating procedures (SOPs), and instructions written in natural language, then prepare workflows and skills for the help desk, as well as forms and policies for access management, employee onboarding and offboarding journeys, and dashboards. Users can also upload a documented procedure or spreadsheet and request that it be converted into an executable system, with the result remaining in draft form for review before publication.
From a Text Request to an Operable Workflow
In a demonstration presented by Jake Stauch, co-founder and CEO of Serval, Catalyst created password-reset workflows after identifying connected systems such as Okta, Google Workspace, and Microsoft Entra. It also generated the basic TypeScript code for executing the procedures, with the ability to add approvals and restrict the parties authorized to run the workflow.
Serval says Catalyst is designed to become the primary interface for configuring the platform, so that in the future everything available to administrators through the traditional interface can be carried out through it. The platform also supports investigating operational problems and correcting workflow failures.
Agents That Work Before a Ticket Is Opened
The most distinctive part of Catalyst is its background agents, which operate on a schedule across connected systems and connect multiple signals to look for emerging problems and suggest fixes. In an example provided by Serval, one agent linked network incidents at two offices with switch, DHCP, and ticket-log data, ruled out hardware failures and wireless interference, then concluded that the cause was configuration drift and created a remediation workflow requiring administrator approval.
Why does this announcement matter? The focus here shifts from improving the ticket cycle to reducing the number of tickets themselves. Instead of waiting for a password-reset request, an access problem, or a recurring outage, the platform attempts to turn repetitive work into reusable automation and then continuously look for additional opportunities. However, this direction is not unique; ServiceNow offers tools such as Build Agent and AI Agent Advisor, while Atlassian provides Rovo capabilities, and Freshworks enables the creation of agents through Freddy AI Agent Studio. Serval’s central claim is therefore that it compresses the cycle of discovering, building, and governing an automation opportunity into a single administrative interface, not merely that it can generate a workflow from a sentence.
Models, Governance, and Data
Serval does not develop its own foundation model; instead, it says it tests models from different labs and selects the most suitable one for each task. Stauch said in an interview with Sequoia Capital in May 2026 that OpenAI models were better at user interaction and tool calling, while Anthropic’s Sonnet and Opus models delivered stronger results in code generation. The company’s documentation allows enterprise administrators to enter their own OpenAI or Anthropic keys, including a custom compatible endpoint, but the available materials do not establish that every user can freely choose any model from a self-service list.
Serval says Catalyst inherits the user’s permissions and remains confined to the user’s team workspace, and that everything it creates begins as a draft. An organization can restrict publishing permissions or require formal review and approval before activating automation. Customers retain the rights to their materials and Serval’s outputs under the Master Services Agreement, and the company says it does not use customer inputs or outputs to train its models or those of other parties. Nevertheless, Serval and authorized subprocessors process customer data to operate, support, and protect the service in accordance with the contractual terms.
The service can be run as a SaaS cloud or within the customer’s environment, including a Serval-managed deployment inside an AWS account owned by the customer or a customer-managed deployment on a cloud or on-premises Kubernetes cluster. Pricing is not published and is determined according to deployment size and usage, typically with a single platform fee and an initial trial to estimate requirements.
Usage Indicators and the Next Test
Serval says more than 90% of customers participating in the trial began automation through Catalyst. According to case studies presented by the company, Ramp accelerated workflow construction by 50% and automated the replacement of 600 laptops, saving 150 hours, while Together AI says Serval automates 95% of infrastructure-access requests when needed. These figures are provided by Serval or its customers in case studies and do not represent an independent evaluation.
The most important test will be Catalyst’s ability to operate in large, customized, and complex enterprise environments. Competitors are also developing tools for application creation and discovering automation opportunities, making Serval’s value dependent on how successfully it can connect the operational record, code, permissions, approvals, and proactive agents within a single governable cycle.