JungleGym
Open-source eval playground for autonomous web agents: three datasets behind one API.
About
JungleGym is an open-source playground from the a16z infrastructure team (a16z-infra/JungleGym) for testing and developing autonomous web agents. It is not an agent itself; it is the eval layer agent developers otherwise assemble by hand. It ships three established web-agent datasets behind a unified REST API: Mind2Web (about 2,000 tasks across 137 sites), WebArena (hundreds of tasks on sandboxed sites), and AgentInstruct (about 1,800 chat-style trajectories).
It also bundles TreeVoyager, a tool that handles the DOM-parsing portion of an agent pipeline using an LLM guided by Tree-of-Thoughts and Minecraft Voyager principles. The project is Apache 2.0 licensed, has a couple hundred GitHub stars (community-reported), and was still receiving commits in 2026, making it one of the longer-lived open-source agent eval kits.
The honest picture: this is evaluation and training infrastructure, not a product you deploy. The datasets are academic benchmarks with known limitations (static sites, dated tasks), so treat scores as comparative signal rather than real-world readiness. Status is experimental (v0.9.0 per project docs).
What it can do
- Mind2Web, WebArena, and AgentInstruct behind one API
- TreeVoyager DOM-parsing for agent pipelines
- Cross-dataset agent performance comparison
- Apache 2.0 open-source eval infrastructure
Setup
Clone the repo; query datasets through the hosted REST API or run locally.
Prerequisites
Python environment for local runs; familiarity with web-agent benchmarks.
Use cases
Works with
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 114; Trending score 66/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.