SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI developer platformsagentskit

agentskit

AgentsKit is a toolkit for building AI agents in JavaScript, offering a lightweight core, React and terminal interfaces, an autonomous runtime, tools, skills, memory, RAG and observability capabilities, scalable plug-and-play from a chat interface to a full agent. Its core has zero runtime dependencies and enforces a 10 KB gzip budget via CI. Materials state that it includes 24 focused packages, 7 framework bindings, 25 native adapters, 50 integrations, 140 catalog providers, 5,000+ catalog models, 21 ready-made skills, 17 memory backends and 69 recipes. The project is independently developed by an individual; the core @agentskit/core has reached v1.12.8 with its API frozen at the minor version level, while the rest of the ecosystem is released on independent beta tracks with clearly defined stability tiers.

Open website

About this tool

Purpose

AgentsKit is aimed at JavaScript developers who need to build real AI agents. Its positioning is not another framework, but the "missing toolkit": small, contract-bound, composable. You can start with a single package and gradually scale to a full stack, using plain JavaScript throughout. It primarily solves the current problem of having to piece together multiple incompatible libraries when building agents in JavaScript.

Main capabilities

  • Lightweight core: the core has zero runtime dependencies, with a CI-enforced 10 KB gzip budget.
  • 24 focused packages, zero lock-in, and six formal contracts that make adapters, tools, skills, memory, retrievers and the runtime all replaceable.
  • Provides React and terminal (Ink) UIs, supporting streaming chat, keyboard input and real tool calls.
  • Autonomous runtime, tool registry, memory, RAG and observability hooks.
  • Supports multi-agent delegation: a planner decomposes tasks, and researchers and coders execute their respective parts; delegation is done through tools the model can already call, with no special syntax to learn.
  • Provides CLI initialization: npx @agentskit/cli init.
  • Provides llms.txt and doc-bridge.config.json for agents, with the public API reference under 2,000 tokens.

Usage

The materials only give brief example pointers and do not provide complete installation commands or code details. It is known that the quick start requires no account, API key or network request to run the first agent, and afterwards supported providers can be connected through the same adapter contract. For complete installation and code, see the official documentation and repository.

Applicable scenarios

Suitable for building various forms from chat UIs to full agents; you can use a single package as needed, or combine them without glue code. Not suitable for the following cases: needing only a single OpenAI streaming call; releasing a consumer chat SDK; requiring Python; requiring enterprise-grade observability right now; needing all packages frozen today.

Limitations and maturity

@agentskit/core is v1.12.8, while the rest of the ecosystem is still graduating package by package on a beta track. AgentsKit targets Node.js 20+ and modern JavaScript runtimes, and packages provide strict TypeScript declarations and dual ESM/CJS output, unless platform binding documentation specifies a narrower target. Browser, edge, React Native, Angular, Vue, Svelte, Solid, React and terminal compatibility varies by package, and the README and guides for each package must be consulted. The observability layer is still young, and mature integrations are currently mainly LangSmith/Arize/Helicone. The project is independently developed by an individual and is MIT licensed.

Related tools

Browse category