AgentDock Core Documentation

Technology Stack

This document outlines the core technologies used in AgentDock Core and its reference implementation.

Core Framework

  • TypeScript: Primary language, providing type safety and modern JavaScript features
  • Node.js: Server-side runtime (>= 20.11.0)
  • Vercel AI SDK: For LLM provider integrations

State Management

  • Zod: Schema validation and runtime type checking
  • Zustand: UI state management in the reference implementation

Development Environment

  • pnpm: Package manager (>= 9.15.0)
  • ESLint & TypeScript-ESLint: Code quality and type checking
  • Jest: Testing framework
  • Husky: Git hooks for code quality

Reference Implementation

For details about the open source client implementation built with Next.js, see Next.js Implementation.

Key Technologies

  • Next.js: React framework with App Router
  • React: UI library (^18.2.0)
  • Shadcn/ui & Radix UI: Component primitives
  • Tailwind CSS: Utility-first styling
  • React Hook Form: Form handling
  • React Markdown: Documentation rendering

AgentDock Core

  • Language: TypeScript
    • Provides strong typing, compile-time checks, and improved maintainability for the core library.
  • Runtime: Node.js (LTS versions recommended)
    • The primary execution environment for the core logic when run server-side.
  • LLM Interaction: Vercel AI SDK (ai package)
    • Provides a unified, stream-first interface to interact with various LLM providers (OpenAI, Anthropic, Google Gemini, Groq, etc.).
    • Handles the complexities of different provider APIs for text generation, streaming, and function/tool calling.
  • Storage (Defaults & Included):
    • In-Memory: Default key-value store for sessions/orchestration if no external provider is configured.
    • Redis (@upstash/redis): Included provider for persistent key-value storage, often used with Docker for development.
    • Vercel KV (@vercel/kv): Included provider for Vercel's key-value store.
  • Schema Validation (for Tools): Zod
    • Used within tool definitions to define and validate input parameters expected from the LLM, ensuring type safety and providing clear error messages.
  • Logging: Custom logger utilities (agentdock-core/src/logging)
    • Provides structured logging capabilities within the core library.
  • Package Manager: pnpm
    • Used for managing dependencies and ensuring efficient installation. AgentDock Core will be released as a separate NPM package when it's ready for release.

Open Source Client (Reference Implementation)

This web application demonstrates how to use AgentDock Core.

  • Framework: Next.js (App Router)
    • Provides the foundation for the web application, including routing, server components, client components, and API routes.
  • UI Components: Shadcn/ui & Radix UI
    • Used for building the user interface components (buttons, inputs, layout, etc.). Built on top of Tailwind CSS.
  • Styling: Tailwind CSS
    • A utility-first CSS framework for styling the application.
  • State Management (UI): Primarily React state/context, potentially Zustand for more complex global state if needed.
  • Client-Side Storage: localStorage (potentially secured via SecureStorage from Core for sensitive items like API keys).

Development & Build Tools

  • Package Management: pnpm
    • Manages dependencies with efficient node_modules structure and consistent installs.
  • Task Running/Scripting: pnpm scripts (defined in package.json)
    • Used for build, test, dev, linting, and other development tasks.
  • Testing: Vitest
    • Used for running unit and integration tests.
  • Linting/Formatting: ESLint & Prettier
    • Ensures code quality and consistent formatting.

Optional Backend Services (for Development)

  • Redis: (As mentioned in Core) Typically run via Docker Compose (docker-compose.yaml) for persistent session/orchestration state during development.
  • Redis Commander: A simple web UI (included in docker-compose.yaml) for inspecting data stored in the development Redis instance.

Extended Vercel AI SDK Capabilities

AgentDock extends the Vercel AI SDK with enhanced capabilities:

  • AgentDockStreamResult: Extends the standard StreamTextResult with:

    • Orchestration state tracking
    • Enhanced error handling
    • Custom response transformations

  • LLMOrchestrationService: Bridges the SDK's streaming capabilities with orchestration:

    • Automatically updates token usage in session state
    • Tracks tools used during conversations
    • Provides this state to orchestration rules

For more details, see: