TrikHub
TrikHub/Docs

What is TrikHub?

TrikHub is an open-source registry and framework for distributing secure, optimized AI Agents and Tools (called Triks).

The Problems

Some of the classic problems of utilizing community code within AI agents:

1. Security

AI agents are vulnerable to prompt injection  when external data contains malicious instructions that hijack the agent’s behavior. Traditional defenses (filtering, sandboxing) don’t work reliably.

2. Cost & Efficiency

When agents rely on dozens of micro-tools, API documentation, often conflicting multiple solutions for the same problem, they waste tokens figuring things out through trial and error. A simple task like “download the latest video from this YouTube channel” might require the agent to:

  • Discover the right APIs
  • Read documentation
  • Handle authentication
  • Debug failures
  • Retry with different approaches

This exploratory process is slow and expensive.

3. Distribution

Even with multiple Open Source tools, the landscape lacks a centralized, uniform aggregator and framework to easily find and distribute AI Skills for other Agentic applications to consume.

The Solutions

Optimized Skills

Instead of micro-tools, Triks are meant to be complete solutions to specific problems. Someone who knows how to download YouTube videos creates a Trik with the optimal flow - tested, refined, and token-efficient. Your agent just calls it.

Just as there is an app, or a saas to solve something, there should also be a plug & play opensource agent.

Security by Design

Every Trik enforces Type-Directed Privilege Separation (TDPS):

  • Constrained output types - Tool results are defined with strict schemas (enums, integers, booleans, bounded strings) so the agent only ever sees safe, structured data
  • Output templates - Natural language responses are assembled from validated fields, not raw LLM output or untrusted content
  • No free-form strings - Untrusted content cannot reach the main agent’s context because output schemas reject unconstrained types

The agent makes decisions on safe, typed data. Malicious instructions embedded in external content have no path into the agent’s reasoning.

Learn about Security

Open Source Distribution

Like npm for AI Skills:

# A trik that enables trik management through conversational Agents
trik install @molefas/trikster
 
# Demo Conversational Trik to showcase storage capabilities
trik install @molefas/trik-demo-notes
 
# Demo Tool-mode Trik
trik install @molefas/trik-hash
 
# Demo Conversational Trik to showcase isolation capabilities through containerization
trik install @molefas/site-builder
 
# Demo Conversational Trik to showcase full app
trik install @molefas/ghost-writer
  • Auditable - Inspect any trik’s code before using it
  • Forkable - Clone, modify, improve, republish
  • Community-driven - Anyone can create and share triks
  • No vendor lock-in - Self-host the registry if needed

The true power of open source applied to AI capabilities.

Get Started

Installation Guide

Next Steps