AI agents in JS, built by someone who's shipped for 15 years.
Software Architect. 180+ engineers led. 500+ engineering interviews. Now building AgentsKit — the most complete JS ecosystem for AI agents.
15+
Years Experience
180+
Engineers Led
500+
Interviews
Trusted by devs shipping agents in production. No fluff.
About Emerson Braun
Software Architect with 15+ years shipping production code at startups and enterprises across LATAM. Led 180+ engineers and conducted 500+ technical interviews — I've seen what separates teams that ship AI from teams that just demo it.
Now I build the JavaScript stack for production AI agents. AgentsKit is the flagship: a single ecosystem covering chat UI, runtime, tools, memory, RAG, and observability — no gluing 8 libraries together. Alongside it I maintain AI Summary Hub (145+ curated articles) and Agent Skills (21 production-grade skills for Claude Code, Copilot, and Codex).
Everything is open source, type-safe, and built from production scars — not demo apps. If you're shipping AI agents in JS, this is the toolkit.
AgentsKit
Ship AI agents in JavaScript. Without gluing 8 libraries together.
Chat UI, runtime, tools, memory, RAG, and production guardrails in one ecosystem. Swap OpenAI for Claude, React for terminal, in-memory for vector DB. Start small, grow into the full stack, keep your code intact.
One ecosystem, not eight
Most JS agent code is glue: one library for tool calling, another for memory, a third for RAG, a fourth for tracing. AgentsKit is a single ecosystem with a unified contract — small, composable, provider-agnostic.
- • Pure TypeScript. Browser, Node, Deno, Bun.
- • 5 KB core. Zero dependencies.
- • 7 UI bindings sharing one ChatReturn contract.
- • Provider routing, ensemble, and fallback built-in.
- • Built-in observability, eval, and sandbox.
# Scaffold a new project — 4 templates, hot reload, demo provider
npx @agentskit/cli init
# Or install the primitives directly
npm install @agentskit/core @agentskit/adaptersIncludes 4 templates, zero-config demo provider, and hot reload via agentskit dev.
The package ecosystem
@agentskit/core
Zero-dependency base layer. Contracts, chat controller, and primitives. 5 KB substrate, runs in browser, Node, Deno, and Bun.
@agentskit/adapters
Provider adapters with routing, ensemble, and fallback. Swap OpenAI for Claude without touching your code.
@agentskit/runtime
ReAct loop implementation, durable execution, and background agent support.
@agentskit/tools
Built-in tools plus MCP bridge integration. Type-safe schemas, predictable failure modes.
@agentskit/memory
Chat, vector, graph, and encrypted stores. Pluggable backends — bring your own.
@agentskit/rag
Ingestion, retrieval, reranking, and document loaders. No magic, just primitives.
@agentskit/skills
Personas and marketplace. Reusable agent capabilities.
@agentskit/observability
Trace viewer, audit logging, and devtools. See what your agent actually did.
@agentskit/eval
Test suites with replay and snapshot diffing.
@agentskit/sandbox
Secure code execution with policy controls.
@agentskit/cli
Init, chat, execute, diagnose. `npx @agentskit/cli init` scaffolds in seconds.
Also building
AI Summary Hub
Curated learning paths for AI engineering
A growing library of in-depth summaries on AI fundamentals, model architectures, agent patterns, and production engineering. Bilingual (EN / PT-BR), structured for both quick reference and deep dives.
Agent Skills
Production-grade skills for AI coding agents
A library of reusable, battle-tested skills for Claude Code, GitHub Copilot, and Codex. Each skill is focused, documented, and ready to drop into your workflow — from code review to automated refactors.
Cracking the Technical Interview
The classic interview prep guide — data structures, system design, behavioral rounds, negotiation. Now free for the whole developer community. When you're ready for the new interview layer, the AI Era Edition takes you one step beyond.