Skip to content
Cursos/DeepSeek Harness (DSH)/DeepSeek Harness · Phase 1
Lección
Revisión pendiente desde 21 ago 20262 fuentes primarias

What is DeepSeek Harness and why does it matter?

DeepSeek Harness (DSH) is an execution layer that turns a language model into an agent able to read and edit files, use tools, keep sessions and request approvals. This ten-phase series goes from the first safe run to local, auditable workflows.

Misión de la lección

Qué vas a conseguir ahora

Una decisión o prueba aplicada a «What is DeepSeek Harness and why does it matter?».

  1. 1Entiende el criterio
  2. 2Haz una práctica pequeña
  3. 3Guarda una evidencia
Evidencia

Una nota breve con qué hiciste, qué salió bien, qué falló y qué revisarías después.

Cuidado. DSH is early-stage software. Its interface, packages and model names may change. Always check the official documentation and repository before using commands in production. Community performance figures are indicative, not guarantees.
Objetivos de aprendizaje
  • Understand the difference between a model and the harness around it.
  • Understand plugins, profiles, sessions, workspaces and subagents.
  • Understand why Cordis, modes and the Trajectory are the differentiating layer.

The formula: model + harness

The model provides language understanding and generation. The harness provides the operating body: tools, permissions, session memory, interface, action logs and a loop that decides when to call each tool.

Terminal
Agent = Model + Harness

Model: reasoning and generation
Harness: tools, permissions, sessions and traceability

Everything is a plugin

DSH is built around the idea that models, tools, skills, sessions, storage and the interface can be swapped as components. This makes it possible to test providers and build specialized profiles without rewriting the whole agent.

  • Profile: a composable configuration of plugins and permissions.
  • Session: a record of prompts, decisions, calls and results.
  • Workspace: the explicit folder where the agent is allowed to work.
  • Skills: reusable instructions for focused tasks.
  • Subagents: secondary agents that split a complex mission.
Idea clave. The value is not that DSH magically makes a model smarter. The value is that it makes the surrounding system configurable and visible: what the agent can touch, which tools it uses, what gets logged and when it must ask for approval.

Why it matters

DSH is not a local-model tutorial. It is a way to turn a DeepSeek model into an observable agent: execution modes, permissions, plugins, profiles, skills, subagents and a Trajectory that lets you review what happened.

This series focuses on the harness. When local models appear, they appear only from the DSH perspective: which provider to select, how to measure and when the official API is still the better choice.

The ten-phase roadmap

  1. Introduction and architecture.
  2. Installation and a read-only first session.
  3. DeepSeek models, providers and cost.
  4. Standard, Code/PTC, Minimal and Creator modes.
  5. Workspaces, permissions and security.
  6. Plugins and custom profiles.
  7. Subagents, skills and orchestration.
  8. Local DeepSeek models and quantization.
  9. Use cases, metrics and benchmarks.
  10. Troubleshooting and production.
Comprueba que funciona. In your own words, explain which part is the model and which part is the harness. Then define a test workspace and write down one action that should require human approval.
Guardar y reabrir el proyecto.
Treat DSH as a configurable, auditable agent architecture, not as a magical clone of another tool. That distinction will help you choose models, plugins and permissions with care.
Siguiente misión recomendada

Si has guardado la evidencia de esta lección, continúa con «Installation and a safe first session». Si no, repite la comprobación antes de avanzar.

Installation and a safe first session
Complete Aulafy mapSee how this lesson fits without leaving your path.

Complete Aulafy map

How all courses connect

This is not a checklist. Start with the foundation, choose an outcome, and go deeper only when your project needs more control.

  1. 1Understand
  2. 2Apply or build
  3. 3Operate with confidence
01

Choose an application

Turn the foundation into a visible outcome: a website, a business improvement, media, or an interactive experience.

AI for small businesses and freelancersUse AI where it improves real work.Beginner to intermediate · 20 lessonsView course Generative AI: image, voice and videoCreate production-ready media with AI pipelines.Beginner to intermediate · 7 lessonsView course Games, 3D and simulations with AIPrototype interactive worlds with AI-assisted tools.Beginner to intermediate · 6 lessonsView course

Continue into the technical branch when you need to maintain code, data, or infrastructure.

02

Build with code

Prepare your environment, work with coding agents, and run models while keeping control of your projects.

Aulafy foundationsPython, Git, terminal, and Docker for learning AI smoothly.Beginner · 6 lessonsView course Codex for programmersBuild, test, and ship changes with OpenAI Codex.Beginner to advanced · 10 lessonsView course Claude Code, from zero to proBuild real software by talking to AI in your terminal.Beginner to advanced · 24 lessonsView course Claude Code + Local AIPrivate AI tools that run on your machine.Intermediate · 34 lessonsView course

This branch prepares you to design and operate reliable AI systems.

03

Take systems to production

Combine retrieval, agents, evaluation, security, deployment, and model adaptation when the problem requires it.

Advanced and secure RAGBuild retrieval systems that answer better and leak less.Intermediate · 14 lessonsView course Agents and automationDesign agents that can work, recover and stay under control.Intermediate to advanced · 20 lessonsView course Production agents with LangGraph and n8nTake agent prototypes into reliable operations.Intermediate to advanced · 9 lessonsView course Self-hosted AI automation for small businessesRun AI workflows on your own infrastructure.Intermediate · 10 lessonsView course AI Router and content systemRoute local and frontier models with quality, cost, and privacy controls.Intermediate to advanced · 6 lessonsView course AI security and evaluationMeasure, test and harden AI systems before trusting them.Intermediate · 7 lessonsView course Local MLOps and model deploymentServe, monitor and operate models from your own stack.Intermediate to advanced · 10 lessonsView course Fine-tuning and post-training for LLMsTrain useful adapters without losing control.Intermediate to advanced · 8 lessonsView course

You do not need every course: choose the component your system needs and return as it grows.

View full catalogue