首页

投稿

[中配] DeepSeek Harness：用免费模型构建AI编程Agent

377

0

2026-10-24 17:01:21

DeepSeek Harness 开源框架，让免费模型拥有顶级 Agent 能力！本课程带你深入掌握这一轻量高效的智能体框架，从核心架构、工具链编排到多 Agent 协作，亲手用 DeepSeek 模型搭建能自主写码、调试、执行任务的 AI Coding Agent。零成本、全实战，摸透 Harness 设计思想。

DeepSeek Harness

deepseekharness

服务器守夜人发消息

在互联网大厂努力扒活ing

001 Welcome and course outcome - 欢迎与课程目标

03:16

002 Model versus agent versus harness - 模型、agent、harness 的区别

02:52

003 How the zero-cost architecture works - 零成本架构原理

02:56

004 Developer preview, telemetry and safety - 开发者预览、遥测与安全

03:31

005 Development-environment requirements - 开发环境要求

02:44

006 Install Node.js and verify the environment - 安装并验证 Node.js

02:12

007 Start DeepSeek Harness with NPX - 用 NPX 启动 DeepSeek Harness

02:05

008 Create a safe practice workspace - 创建安全练习工作区

02:07

009 Navigate the DeepSeek Harness Web UI - 浏览 DeepSeek Harness Web UI

01:45

010 Diagnose installation and startup failures - 诊断安装与启动失败

02:38

011 Choosing a provider that can run an agent - 选择可运行 agent 的提供商

03:01

012 Create a Google AI Studio API key - 创建 Google AI Studio API Key

03:01

013 Configure Gemini in DeepSeek Harness - 在 DeepSeek Harness 中配置 Gemini

02:04

014 Run your first tool-enabled session - 运行首个支持工具调用的会话

02:54

015 Gemini free-tier limits and privacy - Gemini 免费额度与隐私

02:25

016 Troubleshoot Gemini connections - 排查 Gemini 连接问题

02:39

017 Why add a second model provider - 为何添加第二个模型提供商

02:23

018 Create and secure an OpenRouter API key - 创建并保护 OpenRouter API Key

02:35

019 Configure OpenRouter in DeepSeek Harness - 在 DeepSeek Harness 中配置 OpenRouter

01:49

020 Select free coding-agent models - 选择免费的编码 agent 模型

02:22

021 Why not rely on automatic free routing - 为何不依赖自动免费路由

02:22

022 Run the same task with Gemini and OpenRouter - 用 Gemini 和 OpenRouter 执行相同任务

02:38

023 Give the agent a precise coding objective - 为 agent 设定精确编码目标

02:38

024 Repository exploration with read-only tools - 使用只读工具探索仓库

02:03

025 Use Plan Mode before implementation - 实现前先使用计划模式

02:15

026 Approve controlled file modifications - 批准受控的文件修改

02:23

027 Run tests and collect evidence - 运行测试并收集证据

01:43

028 Continue after a failed test - 测试失败后继续执行

02:39

029 Inspect the complete trajectory - 查看完整轨迹

02:34

030 Persistent project instructions - 持久化项目指令

03:01

031 Build a reusable repository-inspection skill - 构建可复用的仓库检查技能

02:44

032 Create a custom command - 创建自定义命令

02:36

033 Understand the plugin architecture - 理解插件架构

02:31

034 Build a simple custom tool - 构建简单自定义工具

02:22

035 Package and reuse the plugin - 打包并复用插件

02:12

036 When to use one agent or several - 何时使用单个或多个agent

02:30

037 Delegate repository analysis to a child agent - 将仓库分析委派给子agent

02:22

038 Build an implementation-and-review workflow - 构建实现与审查工作流

02:16

039 Run DeepSeek Harness in headless mode - 以无头模式运行DeepSeek Harness

02:25

040 Use the Python SDK - 使用Python SDK

02:32

041 Design a repeatable maintenance workflow - 设计可重复的维护工作流

02:31

042 Capstone requirements and acceptance criteria - 结课要求与验收标准

02:12

043 Phase 1 — Inspect and diagnose - 阶段一：检查与诊断

01:59

044 Phase 2 — Implement under approval controls - 阶段二：在审批控制下实施

02:28

045 Phase 3 — Test and recover from failure - 阶段三：测试并从失败恢复

02:12

046 Phase 4 — Produce an auditable completion report - 阶段四：生成可审计完成报告

02:48

047 Cloud inference versus fully local inference - 云端推理与完全本地推理

02:31

048 Install and test Ollama - 安装并测试Ollama

01:41

049 Connect Ollama to DeepSeek Harness - 连接Ollama到DeepSeek Harness

02:49

050 Test a small local model - 测试小型本地模型

02:53

051 Understand local-model limitations - 了解本地模型的限制

02:52

052 Choose the right model route - 选择合适的模型路径

01:55

053 Production-readiness checklist - 生产就绪清单

01:45

054 Final challenge and next steps - 最终挑战与后续步骤

02:25

顶部