LittleHorse Ships Saddle Command Center 1.3 With No-Code Agent Controls, Free Serverless Trial

The privately held company positions Saddle Command Center as an "action layer" that spans artificial intelligence agents, microservices, software-as-a-service applications and event streams. Its approach, which the company calls "Business-as-Code," turns a business process into version-controlled, testable software represented as a graph that technical and business users can inspect and modify.

The platform targets the fragmentation that builds as companies spread processes across specialized applications, integrations, spreadsheets and manual procedures. LittleHorse founder and Chief Executive Colt McNealy said most enterprises run dozens of SaaS applications stitched together by manual integrations. That fragmentation makes agents difficult to adopt for two reasons, he said: "One is lack of context. Two is the difficulty of getting the right things to happen at the right time."

Version 1.3 adds an Apache Kafka-compatible streaming engine, connectors for webhooks and data sources, and a runtime for prebuilt agents and task workers. Developers can convert existing code into workflow components and connect systems through application programming interfaces, event queues or custom tasks.

The release's agent controls follow what LittleHorse calls the decision worker pattern. An agent receives read-only access to tools and handles a task requiring judgment, while deterministic workflow code validates its output and carries out the action, keeping authority out of the agent itself.

"What our customers started doing early on is reducing the scope of the agent to just the fuzzy tasks and then having the workflow engine orchestrate the difficult parts," McNealy said. Without that visibility, he said, "your token bill goes up, your reliability goes down, and you get no observability of what's going on."

Users configure an agent by choosing Model Context Protocol servers, defining a system prompt and specifying input and output structures. Saddle Command Center then deploys the model as a task worker that can be inserted into a larger workflow. The platform records tool calls and provides a stack trace for debugging. Business errors follow rules defined in the workflow, while technical errors such as network failures can be retried according to policy.

The system is also intended for long-running processes that cross organizational and application boundaries. McNealy cited a European payments provider whose merchant-onboarding process involved multiple identity services and responses that could take as long as three days. Using LittleHorse, the provider codified error-handling and timeout procedures for calls to third-party systems and automated workflows based on the responses it received; LittleHorse coordinated the steps, waits and error handling without replacing any underlying systems. "They were able to remove a ton of manual work and speed up their onboarding time with this workflow," McNealy said.

Saddle Command Center supports Java, Python, Go and C#, and version 1.3 adds JavaScript. The free serverless option lets developers test the platform without provisioning infrastructure.

LittleHorse was founded in 2022 by McNealy, whose father, Sun Microsystems Inc. co-founder and former CEO Scott McNealy, is an investor and adviser. The company uses an open-core model, with an open-source orchestration kernel supporting the commercial platform. LittleHorse said its software has been in production for more than 18 months, has paying customers across multiple industries and expects to book more than $1 million in revenue this year. Enterprise sales often begin with a three-week Business-as-Code workshop, while software companies more commonly adopt the open-source version before buying support or hosted services. Enterprise pricing is based on processing capacity, including Kafka throughput and workflow tasks per second.

Saddle Command Center 1.3 is available now.