What is AutoGPT?
AutoGPT is an AI agent platform from Significant Gravitas that turns plain-English requests into agents that carry out complete workflows. The hosted version gives you a team of ready-made AI Experts for jobs such as SEO content, sales outreach and executive assistance, and the homepage counted 33 Experts in October 2026. A coordinating agent called Otto, described as your Head of AI, takes each request, brings in the right Experts and reports back. Developers can instead open the Build canvas, connect blocks step by step and run agents on demand, on a schedule or from a trigger.
AutoGPT began as an experimental autonomous agent released on 30 March 2023 by Toran Bruce Richards, founder of Significant Gravitas Ltd, and it became the top trending repository on GitHub. Wikipedia reports the company raised $12 million in October 2023. The GitHub repository showed about 187,600 stars and 45,900 forks when we checked in October 2026. The original agent lives on as AutoGPT Classic under the MIT licence, while the newer platform code uses the Polyform Shield licence and is sold as a hosted subscription with a free self-hosted option.
Key Features
- Otto and AI Experts: AutoGPT gives you a team of named AI Experts, such as an SEO Content Manager, an Account Executive and an Executive Assistant, led by a coordinating agent called Otto. You describe the outcome once, and Otto decides which Experts should work on it.
- Custom Experts: AutoGPT lets you create your own Expert when the ready-made team does not fit your process. You brief it in plain language, give it access only to the accounts it needs and keep it alongside the built-in specialists.
- AutoPilot chat-to-agent: AutoGPT’s AutoPilot turns a plain-English conversation into a working agent without a workflow canvas. This suits non-technical users who would rather explain a job than wire up nodes, and the agent can be refined later in the builder.
- Visual Build canvas: AutoGPT includes a block-based canvas for defining each step of an agent by hand. You can connect, inspect and test blocks individually, which gives developers more control than the chat route when a workflow must run the same way every time.
- Routines and triggers: AutoGPT runs agents on demand, on a schedule or from a trigger, so recurring work runs without a prompt. The homepage example is a weekly brief delivered at 9:00 AM every Monday.
- Per-Expert permissions and spending caps: AutoGPT lets you choose which connected accounts each Expert can use and set a spending approval threshold for each one. Scheduled workflow runs draw on a credit wallet, so the caps stop a busy agent from draining your balance.
- Agents dashboard and activity view: AutoGPT’s Agents dashboard shows each agent’s runs, costs and the actions waiting for your attention. The activity view lets you follow an Expert’s progress, and the platform says work is reviewed before anything is sent.
- Marketplace of community agents: AutoGPT hosts a marketplace of ready-made agents built by the community. You can add one to your library and adjust it rather than starting from a blank canvas, which shortens setup for common tasks.
- 45-plus app connections: AutoGPT connects agents to more than 45 platforms, according to its GitHub page, including Gmail, Google Sheets, Slack, Notion, HubSpot, Jira and Salesforce. The hosted plans also offer access to a wide range of leading AI models.
- Open-source self-hosting: AutoGPT publishes its platform code on GitHub, and you can self-host it with Docker on Linux or macOS. Self-hosters bring their own model API keys and get community support, while Windows users currently follow a manual setup guide.
Pros
- Free self-hosted route: AutoGPT lets you run the full platform on your own servers without a subscription fee. You pay only for infrastructure and model usage, which suits developers and companies that want control over where data is processed.
- Two ways to build: AutoGPT offers both chat-based setup through Otto and AutoPilot and a step-by-step visual builder. Business users can start by describing a goal, and technical staff can open the same agent on the canvas to tighten each step.
- Large community and code base: AutoGPT is one of the most starred AI projects on GitHub, with about 187,600 stars in October 2026. That community supplies marketplace agents, bug reports and tutorials, which helps when official documentation leaves gaps.
- Clear cost controls: AutoGPT’s per-Expert spending caps and approval thresholds address the runaway API bills that made early autonomous agents risky. You can see costs per run in the dashboard and stop an Expert before it overspends.
- Broad integration list: AutoGPT connects to more than 45 tools, from Gmail and Google Docs to GitHub, Linear, Stripe and Webflow. Agents can therefore act inside the apps your team already uses instead of passing text back for manual copying.
- Useful for content and research: AutoGPT is most often praised on G2 for efficient content creation, time savings and organising research. Reviewers there give it 4.4 out of 5 from 48 reviews, with ease of use among the common positives.
Cons
- Hosted plans are not cheap: AutoGPT Pro costs $50 a month and Max $320 a month, and workflow runs are billed separately from a credit wallet. G2 reviewers most often cite high pricing as a drawback, especially when free agent frameworks exist.
- Answers can still be wrong: AutoGPT agents can produce incorrect or inconsistent results, a complaint that appears repeatedly in G2 reviews alongside hallucinations. Check every draft, email or data change before it reaches a customer or a live system.
- Learning curve for custom workflows: AutoGPT’s builder and self-hosted setup require technical knowledge, and G2 reviewers mention a complex interface and setup. Non-technical users will get further with the ready-made Experts than with the canvas.
- Self-hosting takes real work: AutoGPT’s self-hosted version needs a running Docker environment, your own API keys and ongoing patching. The one-command installer was still awaiting release when we checked, and Windows users must follow a manual guide.
- Card needed for the trial: AutoGPT asks for a payment card before the seven-day Pro trial, although nothing is charged on the day. Set a reminder to cancel if you only want to test it, because Pro billing starts when the trial ends.
- Platform licence limits resale: AutoGPT’s platform code uses the Polyform Shield licence rather than a fully permissive one. You can use it personally or inside your business, but you cannot offer it as a competing hosted service.
Who is Using AutoGPT?
- Small business owners: Delegating content, outreach and admin to ready-made AI Experts.
- Marketing and sales teams: Running scheduled SEO drafts, prospect research and weekly briefs.
- Developers and AI engineers: Self-hosting the platform and building custom block-based agents.
- Researchers and hobbyists: Experimenting with autonomous agents through the open-source Classic code.
Pricing
AutoGPT Pro costs $50 a month after a seven-day free trial, Max costs $320 a month, and self-hosting the open-source platform is free (official pricing page, October 2026).
| Plan | Price | What you get |
|---|---|---|
| Self-Host | Free (your own servers and API keys) | Docker self-hosting, unlimited Otto Chat allowance with your own LLM keys, community and GitHub support |
| Pro | $50/month or $510/year | Seven-day trial, Otto and Expert chat, visual workflow builder, connected apps, scheduled routines, spending controls, email support |
| Max | $320/month | Everything in Pro with 8.5 times Pro chat usage, five times file storage, early access, expanded integrations and priority support |
| Team | Contact sales (coming soon) | Multi-user workspaces, admin roles, central billing, approvals and shared integrations |
Scheduled workflow runs are paid separately from a credit wallet on every plan, and the site does not publish wallet rates. CostBench describes AutoGPT as custom-priced with no trial, which conflicts with the official page, so treat the vendor’s checkout as the final word and review the trial terms before entering a card.
What Makes AutoGPT Unique?
Few agent platforms span both ends of the market the way AutoGPT does. The same project gives non-technical buyers a hosted team of named Experts run by one coordinating agent, while developers can download the platform code and run it on their own hardware. Per-Expert permissions and spending caps respond directly to the looping and cost problems that made the 2023 version notorious, which makes the current product far more practical than its origins.
Compatibilities and Integrations
- Platforms: Web app for the hosted plans; self-hosting via Docker on Linux and macOS, with a manual guide for Windows.
- Integrations: Gmail, Google Calendar, Docs, Sheets and Drive, WordPress, GitHub, Slack, Discord, Notion, HubSpot, Linear, Airtable, Jira, Salesforce, Stripe and Webflow.
- AI models: Access to a wide range of leading models on hosted plans; bring your own API keys when self-hosting.
- Developer tools: Open-source code on GitHub, the Forge agent toolkit and the agbenchmark testing package in AutoGPT Classic.
AutoGPT Tutorials
A first working Expert takes about 15 minutes to set up on the hosted trial.
- Sign up at agpt.co and start the seven-day Pro trial, or follow the GitHub guide to self-host with Docker.
- Tell Otto what you need in plain English, such as a weekly SEO article or a list of sales prospects.
- Connect only the accounts each Expert needs, such as Google Drive or WordPress, and set a spending cap.
- Review the first result in the activity view, then switch it into a scheduled routine.
- Open the Build canvas or the marketplace when you need a custom multi-step agent.
How We Rated It
| Criterion | Score | Why |
|---|---|---|
| Accuracy and Reliability | 3.6 | Inconsistent results reported |
| Ease of Use | 3.7 | Chat easy, builder technical |
| Functionality and Features | 4.5 | Experts, builder, marketplace, routines |
| Performance and Speed | 3.8 | Depends on model and load |
| Customization and Flexibility | 4.6 | Custom Experts, open code |
| Data Privacy and Security | 4.0 | Per-Expert permissions, self-hosting |
| Support and Resources | 4.0 | Large community, email support |
| Cost-Efficiency | 3.5 | Free self-host; pricey hosted |
| Integration Capabilities | 4.4 | 45-plus apps, many models |
| Overall Score | 4.0 | Average of the nine criteria |
Summary
AutoGPT suits small businesses that want a hosted team of AI Experts for content, outreach and admin, and developers who want an open-source agent platform they can run themselves. Its main limitations are hosted prices that climb quickly and agent output that still needs checking. We rate it highly for flexibility and integrations, with caution on accuracy and cost. Start with one narrow routine, set a low spending cap and review every result for a week before widening its access.

Reviews
Write a Review