Appearance
Agent-E
Agent-E is an open-source, agent-driven browser automation system. It lets you control a web browser using plain natural language commands. You describe the task; Agent-E plans and executes the steps inside a real browser.
What it does
Agent-E is built on the AG2 agent framework and uses a hierarchical planning approach backed by a large language model (LLM). It supports:
- Filling out web forms with information you supply or scraped from another site
- Searching and sorting products on e-commerce sites (Amazon, etc.)
- Locating specific content on any website (sports scores, contact info, etc.)
- Navigating to and interacting with web-based media (YouTube, full-screen, mute)
- Running comprehensive web searches and returning summarized results
- Managing tasks on project management platforms such as JIRA
- Personal shopping assistance based on user criteria
It ships a FastAPI HTTP wrapper so you can drive it programmatically or integrate it into larger pipelines.
Category
OSS Agent (browser automation / AI web agent)
Free vs Paid
- Open source: MIT License. Free to self-host.
- Managed service: Emergence AI offers a hosted, enterprise-enhanced version with advanced logging, role-based access, and cloud-scalable infrastructure. Free trial available.
- Managed service: https://www.emergence.ai/orchestrator
Official links
- GitHub: https://github.com/EmergenceAI/Agent-E
- Research paper: https://arxiv.org/abs/2407.13032
- Blog: https://www.emergence.ai/blog/distilling-the-web-for-multi-agent-automation
- Discord: https://discord.gg/wgNfmFuqJF
Template / boilerplate
No downloadable starter template is shipped. The repo includes agents_llm_config-example.json and .env-example as configuration scaffolds to copy and fill in.
Safety and ToS notes
- Agent-E operates on live websites. Every action it takes is real: clicks, form submissions, and purchases execute against real services.
- The project deliberately uses "configured skills" to keep behavior safer and more predictable, rather than giving the agent unconstrained access to arbitrary OS actions.
- You must supply your own LLM API key (OpenAI or compatible). Usage against that API is billed by the LLM provider.
- Respect the terms of service of every site you automate against. Automated access may violate site ToS (e.g., Amazon, JIRA, Google).