Serval is launching Catalyst to general availability Thursday. The AI agent sits above Serval's service management platform as a single administrative layer that moves from discovering automation opportunities in ticket history, to generating TypeScript-backed workflows, to spinning up background agents that probe connected systems for problems before any employee files a ticket. In a live demo, CEO Jake Stauch showed Catalyst detecting Okta, Google Workspace, and Microsoft Entra integrations from a single prompt, then generating the code to automate password resets across all three.
The model stack underneath Catalyst is deliberately swappable. Per Stauch's May 2026 Sequoia interview, Serval runs GPT models for end-user interactions and tool calling, while Anthropic Sonnet and Opus handle code generation. Serval runs continuous evals rather than auto-upgrading workloads to each new release. Org admins can now supply their own OpenAI or Anthropic API keys via Serval's settings. The company is not building a foundation model. Its claimed edge is the harness: enterprise memory, integrations, permissions, approvals, and agent governance.
The competitive context is what makes this worth reading in full. ServiceNow already ships Build Agent for natural-language-to-application generation, AI Agent Advisor for scanning operational records, and AI Agent Studio for agentic workflows. Atlassian Rovo generates Jira automation from plain English. Freshworks offers Freddy AI Agent Studio with 30-plus prebuilt templates. Serval's actual argument is not that it has generative AI, but that Catalyst compresses discovery, assembly, and proactive monitoring into one conversational surface, while legacy platforms require multiple tools, specialists, and configuration layers to get from observed problem to production automation.
[READ ORIGINAL →]