A DeepSeek Lançou um Claude Code DE GRAÇA e Sem Cartão #deepseekharness #claudecode #ia #iagratis

DeepSeek Harness Overview (What It Is)

• DeepSeek didn’t launch a new chat model or a new conversational AI (not a “V4”-type model and not just a new Chat feature).
• Instead, it released a local program/tool called Harness that you install on your machine.
• Harness provides an interface to work with AI alongside your files and terminal.
• Key idea from the official project page: “Everything is a plugin.”
  • Harness is framed as a plugin-based ecosystem, not a single monolithic product.

What Problem Harness Solves vs. “Just Using ChatGPT”

The speaker contrasts Harness with opening an AI directly inside a chat app:

• Harness provides an environment composed of plugins, so you can connect the tools you want without being overwhelmed by an all-in-one toolset.
• It’s described as model-agnostic:
  • You don’t have to use DeepSeek specifically—you can connect other providers and models.

Tutorial: Installing and First Run

Prerequisite

• Install Node.js from nodejs.org.

Launch via NPX

• Run Harness using an NPX command (copy/paste into Windows Terminal or your terminal).

API Key Prompt (Optional at First)

• When you open the provided local/hosted link, Harness may ask for a DeepSeek API key.
• The speaker chooses to configure later.

Creating a Workspace (and Why It Initially Fails)

• After launching, you create/select a workspace folder for your project.
• A greeting/chat attempt can fail if there’s no API key configured yet.

Configuring Models via “Providers”

• In Harness settings → Models, add an external provider API key.
• The speaker demonstrates using NVIDIA (described as a free provider):
  • Create an account and generate a key via build.nvidia.com.
• Notes on rate limiting:
  • Example shown: roughly ~40 requests per minute (described as “infinite AI with a limiter within 1 minute”).
• Then:
  • Paste the NVIDIA API key into Harness.
  • Select a model (example: DeepSeek V3 Flash).
  • Confirm the chat works.

Harness UI Features Mentioned

Workflow Controls

• Chat sidebar and common actions:
  • Attach files
  • Compact mode
  • Feedback mode
  • Go mode

Resource / Behavior Controls

• Visibility and controls for:
  • AI/token usage
  • duration/limits
  • read-only vs. full access
  • whether it asks permission
• Chat context limit handling
• Trajectory/history of actions (what happened, who did what, timing)

Project Organization

• A project sidebar similar to VS Code (files, terminal, etc.).

Usage Modes (Different Capabilities)

• Standard
  • Broader capabilities like:
    • editing files
    • running shells
    • web search
    • scheduling
    • skills
    • subagents
• Minimal
  • More restricted—acts mainly as a tool.
• PTC
  • Exposes an SDK so models can combine operations in TypeScript programs.
• Creator
  • Intended for creating your own AI workflows/ways to use it.

Plugins: How the System Works

• Harness settings include pre-prompts (identification/styles for skills/plugins).
• A Plugins tab lists built-in plugin types such as:
  • Shell
  • Agent
  • Loop
  • Sub Agent
  • Web Search
  • (and more)
• The speaker emphasizes modularity:
  • skills and plugin components are configurable
  • plugins can be added/edited

Tutorial: Installing the Plugin Marketplace + Extensions

Unlike VS Code, Harness requires enabling its plugin marketplace:

1. Install PNPM:
   • npm i -g pnpm
2. Install/use DSH
   • described as letting you run Harness tooling without needing NPX each time.
3. Install a marketplace package (example):
   • dsh plugin add profile-web-ads-market
4. Restart Harness:
   • A plugin marketplace UI appears (VS Code-like)
   • browse/discover plugins, themes, install, restart, and enable/disable

Examples of Plugins/Themes Mentioned

• “Better sidebar” plugin theme/extension:
  • improves sidebar capabilities (more controls, git-related options, more subagent-related tools)
• A theme/wallpaper integration example:
  • changes UI appearance (Wallpaper Engine-like)
  • can be deactivated by removing/turning off the installed plugin

Cloud Code Integration + Orchestration / Subagents

• The speaker highlights integration with Cloud Code for developers already using it.
• They show where to configure Cloud Code:
  • session/installer/usage plan
  • Jira integration
• They also emphasize orchestration of sub-agents:
  • one model/plugin can call another
  • forming multi-step agent chains

Practical Warning / Review-Style Analysis

• The speaker repeatedly warns that DeepSeek Harness is still in preview:
  • plugin/version conflicts can break daily use
  • different plugins may require different dependency versions
• Additional caution:
  • plugins are open source and could be malicious
  • install carefully

Speaker / Sources (Main)

Main Speaker

• “M.” (as credited at the end of the transcript)

Primary Referenced Sources

• DeepSeek Harness official project page
• npx / Node.js tooling
• build.nvidia.com (NVIDIA free AI infrastructure + API keys)
• The Harness plugin marketplace (VS Code-like extensions UI)