AgentDock/AgentDock

Build Anything with AI Agents

What it solves

AgentDock provides a framework for building AI agents that balances the creative, non-deterministic nature of LLMs with the need for predictable, reliable system behavior. It allows developers to create sophisticated agents that can execute complex, multi-step tasks without sacrificing control over the execution path.

How it works

AgentDock uses a modular, node-based architecture where all capabilities are implemented as nodes. It consists of two primary components: AgentDock Core (a backend-first, framework-agnostic library) and an Open Source Client (a Next.js reference implementation).

  • Node System: The system is built on BaseNode, with AgentNode handling LLM interactions and tool usage. Custom logic and tools are added as specialized nodes.
  • Configurable Determinism: Developers can decide which parts of a task use LLM inference (non-deterministic) and which follow defined tool execution paths (deterministic). This allows for "hybrid" workflows where an intelligent agent makes a high-level decision to trigger a rigid, predictable sub-workflow.
  • Orchestration: The framework includes an orchestration layer to control agent behavior and tool availability based on the current context.

Who it’s for

It is designed for developers building production-ready AI agents, automated data processing pipelines, and AI-powered applications that require a high degree of reliability and type safety (TypeScript).

Highlights

  • Configurable Determinism: Balance AI creativity with predictable, step-by-step execution.
  • Framework Agnostic: The core library integrates with any Node.js backend stack.
  • Modular Node Architecture: Build complex systems by connecting simple, extensible nodes.
  • BYOK (Bring Your Own Key): Securely use your own API keys for various LLM providers (OpenAI, Anthropic, Gemini, DeepSeek, Groq).
  • Reference Implementation: Includes a complete Next.js client to accelerate development.