Map
↑DeepSeek

Awesome DeepSeek Agent (deepseek-ai curated list)

Wiki summarydeepseekllmcoding-agentreferencedirectory ↳ show in map Markdown
title
Awesome DeepSeek Agent (deepseek-ai curated list)
type
summary
summary
DeepSeek's official curated list of 16 AI agent and coding-assistant tools that ship step-by-step integration guides for DeepSeek-V4-Pro and DeepSeek-V4-Flash
sources
awesome-deepseek-agent
parent
deepseek
tags
deepseek, llm, coding-agent, reference, directory
created
2026-04-30
updated
2026-04-30

awesome-deepseek-agent is a curated list maintained by the official deepseek-ai GitHub organization. Each entry links to a per-tool guide (under docs/) that walks through installing the tool, configuring it to point at DeepSeek's OpenAI-compatible endpoint, and running the first prompt against deepseek-v4-pro or deepseek-v4-flash.

This is a useful reference for two reasons. First, it's the canonical list of agent harnesses DeepSeek considers first-class — so it tells you which tools have been tested against V4's specific behaviors (long thinking traces, expert mode, the trained-in style markers) versus which are merely "OpenAI-API-compatible." Second, the per-tool guides are short and concrete, which makes the list a fast survey of what's currently shipping in the agent / coding-assistant ecosystem.

What's on the list

Mainstream coding agents

  • Claude Code — Anthropic's terminal coding assistant. Pointing it at DeepSeek means using the OpenAI-compatible base URL override.
  • GitHub Copilot and GitHub Copilot CLI — Microsoft's editor and terminal agents. Both support custom-model configuration in 2026.
  • Kilo Code — CLI + editor extension.
  • OpenCode — open-source terminal/web coding assistant.
  • Crush — open-source terminal coding agent with LSP integration and multi-model support.

DeepSeek-aware terminal agents

  • Pi — minimal extensible terminal harness with tree-structured sessions; we track this as pi-coding-agent.
  • Oh My Pi — Pi fork with OMP-specific tools, model roles, MCP, plugins, agent workflows.
  • Reasonix — DeepSeek-native terminal agent with a cache-first loop and MCP-native architecture (the only entry that explicitly says "DeepSeek-native" in its tagline).
  • Deep Code — open-source terminal agent built specifically for V4: deep thinking, reasoning-effort control, agent skills.
  • Langcli — open-source assistant claiming 100% Claude Code compatibility.

Chat-platform agents (China-facing)

  • OpenClaw — open-source personal AI assistant for Feishu, WeChat, extensible via Skills. (Note: this is the actual OpenClaw project. The wiki has build-your-own-openclaw which is a tutorial on building a clone, not this project itself.)
  • AstrBot — agent for Feishu, Telegram, and more, with skills/plugins/MCPs.
  • WorkBuddy / CodeBuddy — agent + coding assistant with custom OpenAI-compatible model config.
  • nanobot — lightweight AI agent with chat-platform integration, memory, and MCP.

Other

  • Hermes — Nous Research's "self-improving AI agent."

What this tells you about V4's intended deployment surface

DeepSeek treats both terminal coding agents and chat-platform agents as first-class targets. The chat-platform group (Feishu, WeChat, Telegram) is China-specific in a way the rest of the AI agent ecosystem isn't — these are the primary work-messaging surfaces in the DeepSeek user base, equivalent in role to Slack in US enterprises. The terminal coding group is internationally familiar, and is where Claude Code, Crush, OpenCode, and Pi already compete.

The "DeepSeek-native" entries (Reasonix, Deep Code, Oh My Pi guide) are doing more than just OpenAI-API plumbing — they're tuning prompts and agent loops around V4's specific traits: long visible <think> blocks, the reasoning-effort knob, and behaviors like the trained-in style markers that aren't part of the OpenAI API surface.

Why the list matters as a reference

The deepseek-ai org doesn't accept just any tool — entries arrive via PR with a working guide that follows the existing structure. So an entry being on the list is a soft signal that:

  • The tool runs against a DeepSeek API endpoint and someone has confirmed the basic agent loop works
  • There's a maintained how-to that won't be stale on day one
  • DeepSeek's own engineers have at least reviewed the guide

This isn't a quality benchmark — it's a "we looked at the guide and accepted it" filter. But for someone surveying the agent ecosystem from a non-OpenAI / non-Anthropic vantage point, it's a useful first stop.

Repo: https://github.com/deepseek-ai/awesome-deepseek-agent

Linked from