本页还没有译文
本页 10 段还没译成简体中文，下面显示的是英文原文。Deep Agents Code
还没翻译译自英文（docs.langchain.com）本页 10 段都还没译
Deep Agents Code (dcode) is an open source coding agent built on the Deep Agents SDK.
It works with any large language model and supports switching providers or models.
Persistent memory carries context across conversations, customizable skills shape behavior, and approval controls gate code execution.
Get started
Run the following command to install Deep Agents Code and launch an interactive session:
curl -LsSf https://langch.in/dcode | bash
dcodeSee the Quickstart to add provider credentials, run your first task, and learn interactive mode.
Capabilities
Run agent tools remotely instead of on your local machine.
Define measurable objectives or grading criteria so the agent can check whether work is done.
Delegate work to task-specific subagents for parallel execution.
Store and retrieve information across sessions, including project conventions and learned patterns.
Summarize older messages and offload originals to storage.
Require human approval for sensitive tool operations.
Extend agent capabilities with custom expertise and instructions.
Load external tools from Model Context Protocol servers.
Trace agent operations in LangSmith for observability and debugging.
Next steps
Install Deep Agents Code, run your first task, and use interactive or non-interactive modes.
Set up credentials, config.toml, environment variables, hooks, and CLI flags.