Do you need to know how to code to use AutoGPT?

Not for AutoPilot mode, where you describe a goal in plain language. The visual canvas (Build) and the dashboard, however, require understanding what an agent actually does, costs and execution logs included. Self-hosting requires Docker and a language-model API key.

Is AutoGPT really open source?

Not in the strict sense today. The original framework, outside the autogpt_platform folder, stays under the MIT license. But the code running the current platform is under the Polyform Shield license, which allows use and modification but explicitly bans turning it into a product that competes with AutoGPT. That's a source-available license, not an open-source license recognized by the OSI.

Does my data leave my infrastructure?

Self-hosted, you keep control of the infrastructure, but calls to the language model still go out to its provider. On the hosted version, everything passes through AutoGPT's servers and its providers, with no region publicly specified. No general commitment not to train on your data is published: only a clause limited to Google Workspace APIs exists.

How much does AutoGPT cost for a small team?

The Pro tier is 42.50 $ per month, billed yearly, for standard individual use. The Max tier, at 272 $/month, multiplies usage volume by 8.5 and adds priority support. A Team tier, built for multi-person collaboration, was announced as "coming soon" at last check. Self-hosting stays free, aside from the cost of the models called.

What's the difference with CrewAI or Relevance AI?

CrewAI stays a code framework, built for developers who orchestrate several agents in Python. Relevance AI targets packaged business agents (prospecting, support) rather than a generic agent. AutoGPT stands out for its four surfaces of a single product: from guided chat to visual canvas, which makes it the most accessible of the three for anyone discovering autonomous agents.

AutoGPT: our review

AutoGPT is an autonomous-agent platform: you describe a goal, and the agent chains the steps to reach it on its own. The project that started the 2023 AI-agent craze has become an enterprise platform, hosted or self-hosted, run out of London.

Cloud or self-hostedPaid·from $42.50/month

Go to AutoGPT's website

What we take away from it

What counts in its favor

Four ways to build an agent (guided chat, visual canvas, dashboard, model marketplace): the technical team and the business team can work on the same tool.
The core of the original framework stays under the MIT license and self-hosts for free, infrastructure and model costs aside.
185,000 GitHub stars and several thousand commits: the project is alive, not an abandoned repository like many of the 2023 open-source agents.

What counts against it

The code running the current platform (the autogpt_platform folder) is under the Polyform Shield license, not MIT: it bans building a product that competes with AutoGPT from it. That's not open source in the OSI sense.
No security certification (SOC 2, ISO 27001) found on the site or in the public documentation, even though the platform runs agents with access to your third-party accounts.
No free plan on the hosted version: the Pro tier starts at 42.50 $/month billed yearly, before even the cost of tokens consumed by the agents.
No mention of a French interface or documentation.

A good choice for a technical team that wants an autonomous agent without writing an in-house framework. Skip it if compliance (health, legal, finance) requires a verifiable certification, or if no one on the team can read an execution log in English.

The facts

First paid tier: $42.50/month — Pro, billed annually
Data hosting: Cloud or self-hosted · source
Categories: AI agents
Data verified on: 30 July 2026

What happens to your data

Is your data used to train their models?: Not verified
Subprocessors: The only clause found on model training only covers data from Google Workspace APIs: "we do not use any data obtained through Google Workspace APIs to train, develop or improve generalised artificial intelligence or machine learning models." Nothing comparable is published for the rest of the data the platform processes, hence "unknown" rather than "no." The vendor, Determinist Ltd (trading as AutoGPT), is an English company based in Altrincham, United Kingdom; the privacy policy mentions "service providers located outside the UK" with no precise region listed.
Retention: "We will not keep your personal data for longer than we need it": no figure is published, the policy pointing to a direct contact for details per data type.

AutoGPT in detail

AutoGPT started the autonomous-agent wave in March 2023: describe a goal in one sentence and let the program break it down, execute and correct the steps leading to it on its own, with no approval needed at every turn. The original script has since grown into a full platform, with a hosted version and a version you install on your own servers.

What it actually does

Four ways to build an agent coexist on the same platform. AutoPilot, the simplest, turns a brief written in plain language into a working agent. The Build canvas lets you compose more precise logic, block by block, for anyone who wants to control each step. The dashboard centralizes running agents, with their status and execution cost. The Marketplace, lastly, offers agents already built by the community or by the vendor, ready to deploy.

A concrete example: an agent that watches a mailbox, extracts quote requests, looks up the rate in a spreadsheet and replies to the customer, with no human prompting a single step between the received email and the reply sent.

Why it interests us

It's one of the few projects offering a free self-hosted version of a full autonomous agent, which connects directly to our AI agent work: keeping control of where the agent runs, and of the keys it holds to act on your behalf, changes the nature of the risk. An agent that has access to your inbox or your CRM has to run somewhere, and that somewhere determines who can see your data go by. Self-hosted, the agent stays with you; hosted by AutoGPT, it passes through their servers, with no region publicly specified in the documentation.

Where it falls short

The project's name maintains a confusion worth clearing up before committing. AutoGPT talks up its open-source roots, but the code that actually runs the current platform is under the Polyform Shield license, not MIT: it allows use and modification, but explicitly bans turning it into a product that competes with AutoGPT. That's not a license recognized by the Open Source Initiative, even though the repository stays public and readable.

Second reservation: no public security certification was found, neither SOC 2 nor ISO 27001, even though the very nature of the product is to give an agent access to third-party accounts. For a regulated profession, that absence weighs more than for internal use with no sensitive data.

Third point: the hosted version has no free tier. You have to pay 42.50 $ per month before even counting the cost of language-model calls, billed separately by usage.

Our recommendation

AutoGPT convinces for a technical team that wants to try an autonomous agent without writing its own framework, and that has the discipline to self-host if the data processed justifies it. For a firm or SMB with no technical lead, or for a sector that requires written certification, it's better to look at a packaged agent with more explicit contractual support.

Who it's for, and who it isn't

Suited for

A product or technical team that wants to prototype an autonomous agent without starting from a blank page
An organization ready to self-host on its own servers to keep control of the data processed
A developer comfortable with Docker and a model API key, for whom the free repository version is enough

Not a fit

A regulated sector (health, legal, finance) that needs a security certification spelled out in writing
A team with no technical lead: the dashboard shows costs and execution logs that someone needs to know how to read
Anyone looking for a turnkey tool in French: everything, from the site to the documentation, is in English

Frequently asked questions

Do you need to know how to code to use AutoGPT?: Not for AutoPilot mode, where you describe a goal in plain language. The visual canvas (Build) and the dashboard, however, require understanding what an agent actually does, costs and execution logs included. Self-hosting requires Docker and a language-model API key.
Is AutoGPT really open source?: Not in the strict sense today. The original framework, outside the autogpt_platform folder, stays under the MIT license. But the code running the current platform is under the Polyform Shield license, which allows use and modification but explicitly bans turning it into a product that competes with AutoGPT. That's a source-available license, not an open-source license recognized by the OSI.
Does my data leave my infrastructure?: Self-hosted, you keep control of the infrastructure, but calls to the language model still go out to its provider. On the hosted version, everything passes through AutoGPT's servers and its providers, with no region publicly specified. No general commitment not to train on your data is published: only a clause limited to Google Workspace APIs exists.
How much does AutoGPT cost for a small team?: The Pro tier is 42.50 $ per month, billed yearly, for standard individual use. The Max tier, at 272 $/month, multiplies usage volume by 8.5 and adds priority support. A Team tier, built for multi-person collaboration, was announced as "coming soon" at last check. Self-hosting stays free, aside from the cost of the models called.
What's the difference with CrewAI or Relevance AI?: CrewAI stays a code framework, built for developers who orchestrate several agents in Python. Relevance AI targets packaged business agents (prospecting, support) rather than a generic agent. AutoGPT stands out for its four surfaces of a single product: from guided chat to visual canvas, which makes it the most accessible of the three for anyone discovering autonomous agents.

Alternatives to AutoGPT

AI agents

CrewAI

Gets several agents to work together on the same task, each with its own role.

Free version, then paid

Official website

AI agents · Automation

Lindy

AI assistant that carries out office tasks (emails, calendar, customer support) via no-code configurable agents.

Paid·from $49/month