vtcode 命令

常用示例

说明

vtcode is an open-source semantic coding agent written in Rust. It connects to one or more LLM providers and exposes a tool-using assistant that can read, edit, and execute code inside a workspace. A Tree-sitter-based parser gives the agent structural understanding of source files in many languages, enabling targeted edits rather than blind text rewrites. Tool execution is gated by a configurable policy: each tool can be set to allow, deny, or prompt, and shell commands are statically validated with tree-sitter-bash before being run. Sandboxing keeps file operations within the configured workspace boundary. Multi-provider support — OpenAI, Anthropic, DeepSeek, Gemini, Z.AI, Moonshot, OpenRouter, MiniMax, Ollama, LM Studio — is wired through a single configuration file with automatic failover, prompt caching, and context-window-aware truncation.

参数

ask _prompt_

Send a single prompt to the configured model and print the response.

exec _command_

Run _command_ through the agent's policy-aware shell runner with tree-sitter-bash validation.

config show|edit|path

Inspect or edit the vtcode.toml configuration.

--provider _name_

Override the default LLM provider (e.g. openai, anthropic, deepseek, gemini, ollama, lmstudio).

--model _id_

Override the default model id for the selected provider.

--workspace _path_

Set the working directory the agent operates in (defaults to $PWD).

--config _file_

Use _file_ instead of the default vtcode.toml.

--no-tools

Disable tool execution; the agent only emits text.

--yes

Auto-approve tool calls instead of prompting per action.

-v, --verbose

Increase log verbosity.

--version

Print the version and exit.

-h, --help

Show usage information.