scootcli

scoot is a terminal coding agent implemented in plain Python. It accepts natural language instructions and can read, search, edit, and run commands in a repository, asking for approval before performing risky operations. The tool communicates with official model APIs including OpenAI, Anthropic, xAI, Groq, and a local Ollama instance, and it has zero third-party dependencies, using only the Python standard library. It is distributed as a single-file zipapp and as a wheel.

The agent is built around a deterministic loop: the model observes the conversation history, decides whether it needs a tool, and repeats until it can answer without calling a tool. scoot provides a command-line interface with a REPL, session management, approval modes, and a provider layer. Models are addressed as provider/model, and a bare name uses the default provider, which is the first provider with a key, otherwise Ollama. Configuration resolves through CLI flags, environment variables, project .env, and global .env files.

scoot includes eight tools: read_file, list_dir, search, write_file, edit_file, run_shell, open_editor, and update_plan. Paths are resolved against the workspace root, and when a tool needs a file outside the workspace, scoot asks for permission. Approval modes can be set to always, auto-read, auto-edits, or yolo. The REPL supports commands such as /help, /status, /model, /approve, /yolo, /worktree, and /sessions. It also supports one-shot prompts, resuming sessions, and machine-readable output with --json.