Browser-based autonomous agent that splits a written goal into tasks and runs them in a loop
AgentGPT lets you name an agent, give it a goal, and watch it break that goal into sub-tasks it then works through in the browser, with starter templates for company research, trip planning and study plans. It came out of the 2023 autonomous-agent wave at Reworkd and the code is open source under GPL-3.0. The hosted version still serves a free tier plus a paid Pro plan, but the repository was archived in January 2026 and the plan page still advertises GPT-3.5-Turbo and GPT-4. Treat it as a working demo of the idea rather than a maintained product.
Free tier is 5 demo agents a day on GPT-3.5-Turbo with limited plugins and web search. Pro is listed at $40/month for 30 agents a day, GPT-4 access and 25 loops per agent, with a custom Enterprise tier above it. Those model names have not been refreshed in years.
Use tool ↗AgentGPT is the quickest way to watch goal-decomposition agents work without installing anything, and the GPL-3.0 source makes it a useful teaching artifact. The catch is that it is frozen in time: the repo was archived in January 2026 and the hosted app is still wired to 2023-era models.
Watch out: The hosted app still loads and still takes payments, but the project is read-only on GitHub and receives no fixes or new model support — check before putting a card on it.
Side-by-side fact sheets against the competitors our research names — pricing, platforms, API and both verdicts on one page.
No reviews yet — be the first to review AgentGPT.
Reviews are tied to your EffectHub account — one review per tool, so the rating for AgentGPT reflects real users.
No questions yet — ask the first one about AgentGPT.
Questions about AgentGPT are tied to your EffectHub account, so answers can reach you and the section stays free of spam.