Natural language agent workflows for browser and Node.js

What it does

Eko is a TypeScript framework that turns natural language into automated workflows. It provides a unified interface for running agents in both browser extensions and Node.js environments, handling tasks like web scraping, file management, and GUI automation. You configure LLM providers and agent types, then invoke the system with a plain string command.

The interesting bit

The framework treats human intervention as a first-class feature: workflows support pause, resume, and interrupt controls with task_snapshot recovery, and recent releases added dependency-aware parallel agent execution. It also claims to access private web resources, though the README does not detail the authentication or sandboxing mechanics.

Key highlights

• Runs as a browser extension, web app, or Node.js process, with Playwright integration shown in the Node.js example.
• Supports multiple LLM providers—Anthropic, Google, OpenAI, and OpenAI-compatible endpoints like Qwen and Doubao—through a unified configuration.
• Native MCP (Model Context Protocol) integration to connect with external tool servers.
• Multi-agent orchestration with dependency-aware parallel execution and human-in-the-loop breakpoints.
• Version 4.0 added chat conversation support and optimized agent logic.

Caveats

• The README explicitly warns against embedding API keys in browser or frontend code, implying web and extension deployments need a careful backend proxy setup.
• Observable Chain and Native A2A are both marked as “Coming soon,” so chain observability and agent-to-agent protocol support remain incomplete.
• The framework comparison table is self-reported and lacks independent benchmarks.

Verdict

Evaluate it if you want one JavaScript codebase to orchestrate browser and desktop automation via natural language. Look elsewhere if you need mature A2A protocols or fully observable reasoning chains today.