Enterprise-Grade Monitoring & Explainability for the Agentic AI Stack
The Complete Observability Stack for LLM Applications — From Prompt to Production
npm install @protoethik-ai/core
# or
pnpm add @protoethik-ai/core
# or
yarn add @protoethik-ai/core| Component | Purpose | Status |
|---|---|---|
| @protoethik-ai/core | Core SDK for intercepting LLM calls | ✅ Production Ready |
| @protoethik-ai/dashboard | Real-time monitoring dashboard | 🚀 Beta |
| @protoethik-ai/ui | React components for embedding | ✅ Stable |
| Providers (OpenAI, Claude, etc.) | Native integrations | ✅ Complete |
In the era of Agentic AI, LLM applications have evolved beyond simple input-output models. The dynamic interactions—models invoking tools, tools feeding back to models, and workflows branching in real-time—have rendered systems into opaque "black boxes."
Tylo-Lens provides Observability, Explainability, and Auditability for LLM systems:
- Invocation Chain Visualization: Tracks the interaction topology across multi-level Agents and Tools.
- Token & Cost Analysis: Real-time computation of consumption per model and per process stage.
- Compliance & Privacy Auditing: Built-in PII (Personally Identifiable Information) scanning and audit report generation.
- MCP Protocol Ready: Native support for the Model Context Protocol introduced by Anthropic.
Tylo-Lens adopts a modular monorepo architecture, ensuring you can integrate only the components you require.
tylo-lens/
├── .github/ # GitHub Actions, Issue Templates, PR Templates
│ ├── workflows/ # CI/CD pipelines (Test, Build, Release)
│ └── linters/ # Custom linting rules for Protoethik standards
├── packages/ # Monorepo workspaces
│ ├── @protoethik-ai/core/ # The "Inspector" - Pure JS logic to intercept MCP/LLM calls
│ │ ├── src/
│ │ │ ├── interceptors/ # Logic for wrapping fetch/XHR/MCP-protocol
│ │ │ ├── collectors/ # Logic for gathering tokens, latency, and costs
│ │ │ ├── ethics/ # PII detection & compliance checking algorithms
│ │ │ └── index.ts # Main entry point for the SDK
│ │ ├── tests/ # Unit tests for the core engine
│ │ ├── package.json
│ │ └── tsconfig.json
│ ├── @protoethik-ai/ui/ # Shared UI components (The "Glass" design system)
│ │ ├── src/
│ │ │ ├── components/ # Atomic components (Nodes, Charts, Traces)
│ │ │ ├── hooks/ # React hooks for real-time data streaming
│ │ │ ├── styles/ # Tailwind config & global CSS (The "Cyber" look)
│ │ │ └── theme/ # Design tokens (Colors, Spacing, Typography)
│ │ └── package.json
│ └── @protoethik-ai/dashboard/ # The main Next.js application
│ ├── src/
│ │ ├── app/ # App Router (Dashboard, Settings, Analytics)
│ │ ├── lib/ # Dashboard-specific logic (Data persistence)
│ │ └── store/ # State management (Zustand/Signals)
│ ├── public/ # High-res assets, Logos, Icons
│ └── package.json
├── examples/ # Integration examples for developers
│ ├── basic-mcp/ # How to use Tylo-Lens with a standard MCP server
│ ├── nextjs-ai-sdk/ # Integration with Vercel AI SDK
│ └── vanilla-js/ # Integration without any framework
├── docs/ # Documentation (Docusaurus or Mintlify source)
│ ├── architecture/ # Deep dive into how Tylo-Lens works
│ ├── compliance/ # Ethics and Transparency standards documentation
│ └── guides/ # Getting started and advanced usage
├── scripts/ # Internal build and maintenance scripts
├── .gitignore
├── .prettierrc
├── .eslintrc.js
├── lerna.json / pnpm-workspace.yaml
├── CHANGELOG.md # Auto-generated by Changesets
├── CONTRIBUTING.md # Guidelines for community
├── LICENSE # Open-source license (e.g., MIT or Apache 2.0)
├── README.md # The masterpiece landing page
└── package.json # Root scripts (build, lint, test-all)
You can install the production-optimized package directly from NPM:
pnpm add @protoethik-ai/coreConfigure the interceptor at your application's entry point to enable automatic data collection:
import {
TyloLens,
networkInstrumentationPlugin,
autoTracePlugin
} from '@protoethik-ai/core';
const lens = new TyloLens({
app: { name: 'my-agent-service', environment: 'production' },
plugins: [
// Automatically intercepts fetch and XHR requests.
networkInstrumentationPlugin({ fetch: true, xhr: true }),
// Manages Trace lifecycle automatically.
autoTracePlugin({ idleMs: 1500 })
],
});
// Your application logic here.To privately deploy the monitoring dashboard, clone the repository and execute the following commands:
# When deploying to Vercel, set the root directory to `packages/@protoethik-ai/dashboard`.
pnpm install
pnpm build
pnpm startTylo-Lens introduces a pioneering, quantifiable transparency scoring system based on the Protoethik framework, assessing key dimensions:
- Clarity: The explicitness of prompts and context.
- PII Risk: The weighted risk of sensitive data exposure.
- Token Efficiency: The resource cost of system operation.
- RFC Process: Any modifications to core data structures must follow the process outlined in the
rfcs/directory. - Security First: The SDK incorporates built-in data sanitization mechanisms, with PII masking enabled by default in production environments.
MIT © 2025 Protoethik
Made with ❤️ by TyloAI
Part of the Protoethik ecosystem
"Building AI that thinks deeply and acts responsibly."
GitHub • TyloAI • Protoethik