Building the Plugin Layer for AI Agents: Exploring the DeepSeek Harness Ecosystem

susu su — Thu, 20 Aug 2026 15:16:59 +0000

AI agents are evolving quickly.

They are no longer limited to answering questions. Modern agents can use tools, access external data, execute workflows, and interact with different services.

However, as agent capabilities grow, a new challenge appears:

How do developers discover and manage the growing number of extensions available for AI agents?

The Rise of AI Agent Plugins

Plugins are becoming an important part of the AI ecosystem.

They allow agents to extend their capabilities without rebuilding the entire system.

Common examples include:

External API integrations
Data retrieval tools
Automation workflows
Memory systems
Developer utilities
Multimodal capabilities

A powerful agent is not only defined by the model behind it, but also by the ecosystem around it.

The Challenge of Finding Useful Plugins

As more AI agent plugins are created, developers face several problems:

Plugins are distributed across different repositories
Installation methods are inconsistent
Maintenance status is unclear
Security and permissions are difficult to evaluate

Finding the right extension often requires manually searching through many sources.

Exploring the DeepSeek Harness Ecosystem

DeepSeek Harness is an emerging AI agent ecosystem that supports plugin-based extensions.

To make plugin discovery easier, I built DSH Marketplace, a searchable platform for exploring DeepSeek Harness plugins.

The platform organizes plugin information including:

Categories
GitHub repositories
Installation methods
License information
Maintenance signals
Related metadata

Currently, it indexes more than 2,500 DeepSeek Harness plugins.

Why Plugin Discovery Infrastructure Matters

Many software ecosystems eventually need discovery platforms.

Examples include:

App stores for mobile applications
Package managers for programming languages
Extension marketplaces for browsers

AI agents will likely need similar infrastructure.

Developers need a reliable way to discover, evaluate, and use extensions.

The Future of AI Agent Ecosystems

In the future, AI agent ecosystems may require:

Standardized plugin metadata
Security verification
Compatibility checks
Better installation workflows
Quality signals from usage data

The next stage of AI development may not only depend on better models, but also better ecosystems around those models.

Project:

https://dshmarketplace.dev

A simple workflow to clean NotebookLM exports before sharing

susu su — Sat, 20 Jun 2026 09:08:37 +0000

When using NotebookLM for research, study notes, tutorials, or presentation materials, the export step is easy to overlook.

A simple workflow I use:

Create or organize the research material in NotebookLM.
Export the material as a video, PDF, PPTX, image, or research document.
Review the exported file before sharing it with students, teammates, clients, or an audience.
Clean the export and remove unwanted watermarks or extra export elements.
Save a cleaner version for presentations, documentation, tutorials, or archiving.

For this step, I use NotebookLM Remover:

https://notebooklmremover.org/

It is a browser-based tool for cleaning NotebookLM exports and removing unwanted watermarks from videos, PDFs, PPTX files, images, and research materials.

This can be useful for students, researchers, educators, content creators, and teams who use NotebookLM as part of their AI research or productivity workflow.