What it does

Dcode, described in the Deep Agents repository as Deep Agents Code, is a ready-made coding agent that runs in your terminal and is powered by the LLM of your choice. The repository gives a one-line shell command to install it. It comes from the Deep Agents project, an open-source agent harness that works out of the box and lets you extend, override or replace any piece. The harness is built on LangGraph and bundles sub-agents with isolated context windows, a filesystem layer with local, sandboxed or remote backends, context management, shell access, persistent memory, human approval of tool calls, reusable skills, and support for your own tools or any MCP server. The project is model-agnostic. Any model that supports tool calling works, whether a frontier API from OpenAI, Anthropic or Google, an open-weight model hosted on a provider such as Baseten or Fireworks, or a self-hosted model through Ollama, vLLM or llama.cpp. The code is released under the MIT license. The project follows a trust-the-LLM security model: the agent can do anything its tools allow, so boundaries should be set at the tool or sandbox level.

Capabilities

Coding agent that runs in the terminal
Works with any model that supports tool calling, including local ones
Sub-agents with isolated context windows (Deep Agents harness)
Human approval of tool calls before they run (Deep Agents harness)
Reusable skills and support for any MCP server (Deep Agents harness)
Persistent memory across sessions (Deep Agents harness)

When to use it

You want a terminal coding agent that is not tied to one model provider
You want to run a coding agent against open-weight or self-hosted models
You want to build on an agent harness you can customize piece by piece