AgentGPT Review
A browser-based autonomous AI agent whose GitHub repo was archived read-only on January 28, 2026, though the free demo still runs on frozen 2023 OpenAI models.
Is AgentGPT Worth It?
AgentGPT was archived read-only by Reworkd on January 28, 2026, so we do not recommend adopting it, use Lindy for no-code business agents or CrewAI for a Python multi-agent framework instead. The hosted demo at agentgpt.reworkd.ai still loads and still grants 5 free runs a day on GPT-3.5, but the project has been frozen since archival at the 2023 OpenAI model lineup, with no patches, no security fixes, and no support. It remains a fine way to watch an autonomous agent loop work in a browser tab for free, but it is not a tool worth building on.
About AgentGPT
AgentGPT is a browser-based autonomous AI agent that Reworkd archived read-only on January 28, 2026, freezing the project at its November 2023 v1.0.0 release with no bug fixes, no security patches, and no support for models released after 2023. AgentGPT has not shut down. The hosted app at agentgpt.reworkd.ai still loads, and it still grants 5 free agent runs a day on GPT-3.5 with zero setup.
The tool takes a written goal, decomposes it into a queue of subtasks, executes each subtask against an OpenAI model plus optional web search, reviews the output, and loops until the goal resolves or a max-loop setting stops the run. Reworkd AI, a Y Combinator S23 company, built and operates AgentGPT, and the open-source repository at reworkd/AgentGPT holds 36.3k GitHub stars, though it now accepts no upstream commits.
Is AgentGPT safe to use, frozen since archival?
AgentGPT is safe enough for low-stakes experimentation but not for sensitive work. It runs on a user-supplied OpenAI API key stored locally in the browser, and because the code has received no security patches since the January 2026 archival, any defect found from that date onward stays open permanently. A scoped API key with a hard spend limit and a low max-loop ceiling are the only real safeguards against a runaway agent.
Our Verdict
Pros
- Zero-setup browser demo still runs an autonomous agent loop for free
- 5 free agent runs per day on GPT-3.5, no API key required
- Open-source code stays public and forkable on GitHub, 36.3k stars
- Visible task decomposition shows each subtask as it executes
- Ready-made templates, ResearchGPT, TravelGPT and StudyGPT, prefill a first goal
Cons
- Archived read-only since January 28, 2026, no patches or security fixes
- Frozen at the 2023 OpenAI model lineup, no GPT-4o, no Claude, no Gemini
- No observability or evaluation tooling for tracking agent runs
- Agents drift off-goal on complex tasks with no correction path
- Paid Pro and Enterprise tiers are listed but dormant and unsupported
How It Works
Open the demo
Go to agentgpt.reworkd.ai in a browser.
Name the agent
Enter a name for the AgentGPT agent.
Set a goal
Enter a goal in one plain sentence.
Cap max loops
Set a low max-loop ceiling in Settings before deploying.
Deploy
Hit Deploy and watch AgentGPT decompose the goal into subtasks.
Watch it iterate
AgentGPT executes, reviews, and queues follow-up subtasks until the goal resolves or the cap trips.
AgentGPT Features
Agent Loop
Decomposes a goal into subtasks and iterates until it resolves.
GPT-3.5 Execution
Runs every free-tier subtask on OpenAI's GPT-3.5 Turbo.
Ready-Made Templates
ResearchGPT, TravelGPT and StudyGPT prefill a goal with one click.
Self-Hosting via Docker
Forks the frozen open-source repo to run locally.
Archived Codebase
Read-only since January 28, 2026, no patches or new models.
AgentGPT Pricing
- 5 agent runs per day
- GPT-3.5 Turbo only
- Templates included
- No API key required
- Listed 30 agents
- Listed 25 loops/agent
- Listed GPT-4 access
- Listed plugins
- Custom agent allowance
- Custom loop ceiling
- Listed GPT-4 access
- No active support
Who Is AgentGPT Best For?
5 free daily runs show an agent loop with zero setup.
A public, forkable repo to study, not a dependency to build on.
No fit, no observability, no patches, no production tooling.
AgentGPT Alternatives
| Tool | Best for | Price | Notes | Compare |
|---|---|---|---|---|
| Lindy | No-code business agents | Not publicly listed | Managed platform | vs → |
| CrewAI | Python multi-agent orchestration | Not publicly listed | Open-source framework | vs → |
| Relevance AI | Managed no-code agent workforce | Not publicly listed | Managed platform | vs → |
| AutoGPT | Open-source local agents | Not publicly listed | Open-source, self-run | vs → |
| AgentGPT — this review | Watching an agent loop in a browser, archived | Free demo, $40/mo Pro listed, dormant | 5 free runs/day, no signup |
AgentGPT vs AutoGPT
AgentGPT runs in a browser tab with zero setup but is archived and frozen at 2023 OpenAI models, while AutoGPT runs locally, takes real setup, and is reportedly still maintained. Zero-setup demo favors AgentGPT, active maintenance favors AutoGPT.
AgentGPT FAQ
Is AgentGPT still working?
Yes, the hosted demo still loads and grants 5 free runs a day, but Reworkd archived the repository read-only on January 28, 2026, so AgentGPT receives no fixes, patches, or new models.
Is AgentGPT discontinued?
AgentGPT is archived rather than discontinued. Reworkd made the repository read-only on January 28, 2026 and pivoted to web scraping, but never announced a shutdown and left the hosted demo running.
Is AgentGPT free?
Yes, the hosted demo gives 5 agent runs a day on GPT-3.5 with no payment method or API key, while the listed $40/month Pro plan sits on archived, unsupported code.
Is AgentGPT worth it?
No, not for adoption in 2026. The repository has been read-only since January 28, 2026 with no patches or new models, so learners can still use the free demo, but teams should choose a maintained platform such as Lindy or CrewAI.
Ready to try AgentGPT?
Start free, or take the trial to explore premium features across all your devices.
Visit AgentGPT →