agentskit: Build AI Agents in JavaScript

Overview

AgentsKit is a modular toolkit for building AI agents in JavaScript and TypeScript. It addresses the work of combining separate libraries for model access, agent execution, tools, memory, retrieval, and interfaces by offering shared contracts across those components.

Developers can adopt individual packages or assemble a broader agent stack, with React and terminal interfaces as well as headless runtime options. It is aimed at people building agents rather than applications that only need a single model call or a chat UI.

Key Features

• Provider adapters for services including OpenAI, Anthropic, Gemini, Ollama, DeepSeek, and Grok.
• An autonomous runtime with tool use, skills, and delegation.
• Tool packages for capabilities such as web search, filesystem access, shell use, and MCP.
• Memory and retrieval packages for chat history, vector search, and RAG workflows.
• UI bindings for React and terminal applications, plus Vue, Svelte, Solid, Angular, and React Native.
• Supporting packages for integrations, sandboxing, evaluation, and observability.
• A small, dependency-free core with documented contracts for substitutable components.

Use Cases

• Build a headless research or coding agent that combines a model provider with tools and skills.
• Add streaming agent chat to a React application, or provide a terminal interface for developer workflows.
• Prototype provider-independent applications where swapping LLM adapters should not require rewriting the rest of the stack.
• Add retrieval and memory to an agent that needs to work with user documents or preserve context across interactions.
• Expose agent tools through MCP for use with compatible clients such as Claude Desktop or Cursor.

Getting Started

Install the core and runtime packages:

npm install @agentskit/core @agentskit/runtime tsx

See the README for a runnable first-agent example, provider setup, and package guides.

Alternatives

• harness-sdk: Strands offers Python and TypeScript agent tooling, while AgentsKit centers on composable TypeScript packages for agents and related app components.
• OxyGent: OxyGent focuses on configurable Python multi-agent systems, while AgentsKit provides a TypeScript toolkit spanning agents, tools, memory, retrieval, and UI.
• deepagents: Deep Agents is a Python harness with planning, delegation, and file access built in; AgentsKit is a TypeScript package toolkit for composing agent applications.

Considerations

• The README identifies @agentskit/core as stable and the rest of the ecosystem as beta, with packages progressing independently. Check the stability policy before relying on beta APIs.
• The project targets Node.js 20+ and modern JavaScript runtimes. Compatibility varies by package and platform binding.
• AgentsKit may be more than needed for a single model request or a straightforward chat SDK use case. The README recommends using a provider SDK directly for a single call and considering a chat-focused SDK for chat-only products.
• The repository describes its observability layer as young, so teams needing mature production monitoring may need additional integrations.
• For details and current package guidance, consult the official documentation.