AgentsKit

What is AgentsKit?

AgentsKit is an open-source JavaScript and TypeScript toolkit for building AI agents. It bundles model adapters, an agent runtime, tools, memory, RAG, observability, and UI bindings into small packages that share formal contracts. The core weighs about 10 KB with zero dependencies, and you can switch model providers by changing one line of code.

Top Features:

• Model adapters: one contract covers OpenAI, Anthropic, Gemini, Ollama, DeepSeek, Grok, and more.
• Agent runtime: run ReAct loops, planning, tool calls, and multi-agent delegation built in.
• Integrations: connect agents to Slack, Notion, Stripe, GitHub, Jira, and dozens more.

Use Cases:

• Research agents: build agents that search the web and save summaries to files.
• In-app chat: add streaming chat with tool calls to an existing React app.
• Document search: answer questions over company files using chunking, embeddings, and retrieval.

Who Can Use AgentsKit?

• JavaScript developers: build production agents without stitching together five separate, competing libraries.
• Startups: prototype quickly and switch model providers later without rewriting app code.
• Platform engineers: trace agent runs with LangSmith or OpenTelemetry and run evaluations.

Pricing

• Core packages (free): every package is MIT licensed and installs from npm at no cost.
• Registry (free): ready-made agents and tools you install as source code and own.
• Model usage (provider rates): token costs depend on your model, while local Ollama models cost nothing.

Pros and Cons

Pros:

• Tiny core: a 10 KB core with zero dependencies keeps app bundles small.
• No lock-in: adapters, tools, memory, and runtimes can be swapped behind shared contracts.
• Quick start: a first agent runs locally with no account or API key.

Cons:

• Solo maintainer: one developer builds the project, which may worry larger teams.
• Code required: there is no visual builder, so non-developers will likely struggle.
• Young ecosystem: there are fewer tutorials and community answers than for LangChain.

FAQs:

1) Is it free?

Yes, the whole toolkit is open source under the MIT license.

2) What runtime does it need?

It is written in TypeScript and runs on Node.js 20 or newer.

3) Can I use local models?

Yes, the Ollama adapter runs local models without needing an API key.

4) Which UI frameworks work?

React, Vue, Svelte, Solid, Angular, React Native, and Ink terminal apps work.

5) How do I start?

Run the CLI init command to scaffold a project with adapters and tools.