Step Code is StepFun’s terminal-based coding agent: it reads code, makes changes, and runs tests in a full autonomous task loop, tuned for token efficiency and long-horizon reliability. The open-source CLI is MIT licensed, has earned over 1,000 stars on GitHub, and is under active development, with updates landing in October 2026.
After signing in, Step Code discovers the available Step models and gets to work in your shell. Long tasks can be split across parallel sub-agents, each with an isolated context, so redundant content never pollutes the main conversation โ a design the team credits for its token savings on extended jobs. MCP servers, agent skills, and plugins work out of the box, and a /goal command lets you delegate long-running tasks for autonomous execution. A distinctive extra is StepPage: a built-in publishing feature that turns a local page into a shareable static website with a single command, complete with version management, closing the loop from development to delivery inside the same terminal session.
Step Code reflects StepFun’s bet that a coding agent should be co-designed with its models: the CLI is tuned alongside the Step model family, so context handling, tool use, and sub-agent delegation are optimized for how those models actually behave. The built-in StepPage publishing is a telling detail โ the workflow assumes an agent’s job is not finished when the code compiles, but when the result is delivered somewhere people can see it, and it folds that last step into the same terminal session.
Step Code itself is free and open source under the MIT license. It runs on StepFun’s Step models, so model usage follows StepFun’s platform pricing after sign-in โ a freemium arrangement.
Step Code targets developers who like terminal-first AI coding tools and want an agent optimized for long, multi-step engineering tasks rather than quick one-shot edits.